
    %Vji             -       n   d dl Z d dlmZ d dlmZ d dlmZ d dlmZm	Z	m
Z
mZ d dlmZ d dlZd dlmZ d dlmZmZmZ d dlmZmZmZmZ d d	lmZ d d
lmZmZ d dlmZm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z) d dl*m+Z+m,Z,m-Z-m.Z.m/Z/ d dl0m1Z1m2Z2 d dl3m4Z5 d dl6m7Z8  e
d          Z9 ed          Z:ej;        j<        Z<ej=        >                    ddd          Z? e@d          \  ZAZBZCdeee:e9f         gee:e9f         f         fdZDde#fdZEd ZFd ZG eDe<jH        e<jI        g           e/            dddejJ        ddfd                        ZK eDe<jL        jM        e<jL        jN        g           e/            d                         ZO eDe<jP        jM        e<jP        jN        g           e/            dd d!                        ZP eDe<jQ                   e/            d"                         ZQ eDe<jR        jM        e<jR        jN        e<jS        jM        e<jS        jN        g           e/d#d$          d%                         ZT eDe<jU        jM        e<jU        jN        g           e/            d&                         ZUd' ZVd-d(ed)eWeX         d*eYfd+ZZ eDe<j[        jM        e<j[        jN        g           e/            d,                         Z\dZ]d)eWeX         fd-Z^ eDe<j_        jM        e<j_        jN        g           e/            d.                         Z` eDe<ja        jb                  dd/d0            Zc eDe<ja        jM                  ejd        dddd1d2            Ze eDe<jf        jM        e<jf        jN        g           e/            ejd        dddd1d3                        Zg eDe<jf        jh        e<jf        ji        g           e/            ejd        dddd1d4                        Zj eDe<jk        jM        e<jk        jN        g           e/            ddddd1d5                        Zl eDe<jm        jM        e<jm        jN        g           e/            d(ed)eWeX         d6eXd7eXfd8                        Zn eDe<jo        jM                  d-d9            Zpd: Zq eDe<jr        jM                  d;             Zs eDe<jt                  	 	 	 d.d<ed=ed>ed?e	e         d@e	eu         dAe	ejv                 fdB            Zw eDe<jx                  	 d/dCedDedEedAe	ejv                 fdF            Zy eDe<jz                  dGdGddHd<edCedDedEedAe	ejv                 f
dI            Z{ eDe<j|                  	 	 	 	 	 	 	 d0dJej        dKej        d?e	e         dLe	e         dAe	ejv                 dMeYdNeXdOeXdPeXfdQ            Z} eDe<j~        jM                  dRdSd(ed)eXdTedUej        dVeudWeYdefdX            Z eDe<j        jM                  dRdSd(ed)eXdTedUej        dVeudWeYdefdY            Z e/             eDe<j        jM                  dZ                         Z eDe<j        jM                  dddd ddd[d\edVeud]e	e         d$e	e         d^e	e         d_eXd`eYdefda            Z eDe<j        jM        e<j        j        g           e/            db                         Z eDe<j        j                  d-dc            Z eDe<j        jM        e<j        j        g           e/            dd                         Z eDe<j        j                  d-de            Z eDe<j        jM                  df             Z eDe<j        jN                  dg             Z eDe<j        jM                  dh             Z eDe<j        j                  di             Z eDe<j        jM                  dj             Z eDe<j        jM                  ddddddkdl            Z eDe<j        jM                  d1dm            Z eDe<j        jM                  d.dn            Z eDe<j        jM                  d1do            Z eDe<j        jM                  dp             Z eDe<j        j                  dq             Zd(edreufdsZd(edtedueufdvZ	 d2dwedreudxeYfdyZd3dtedreudzeufd{Zdted|ed}eYdreufd~Z	 d4deuded<edeufdZdeufdZ eDe<j        jM        e<j        j        g           e/dd          d5dtedeudeYfd                        Z eDe<j        jM        e<j        jN        g           e/            d<edefd                        Z eDe<j        g           e/dd          d<efd                        ZdedefdZ eDe<j                   e/            d(edtedeYdefd                        Z eDe<j                   e/            d-d(edtedeYdefd                        Z eDe<j                   e/            d-d(edeYdefd                        Z eDe<j                   e/            d-d(edeYdefd                        Z eDe<j        jM                  d6dtedeYdeYfd            Z eDe<j        jM        e<j        jN        g           e/            d<ededefd                        Z eDe<j        jM                  d-dtedeYfd            Z eDe<j        jM        e<j        jN        g           e/ddd          dddd(edeYdeYdeeeef         fd                        Z eDe<j        jM        e<j        jN        g           e/            dddeded|edeYdef
d                        Z eDe<j        jM        e<j        jN        g           e/ddd          dRddtedeYdeeeef         fd                        Z eDe<j        jM        e<j        jN        g           e/ddd          dRdddtedeYdeYdeeeef         fd                        Z eDe<j        jM        e<j        jN        g           e/            dRdddeded|ed}eYdeYdefd                        Z eDe<j                   e/ddd          	 	 d7dededeYdeYdeeeef         f
d                        ZdeudeeYeYf         fdZ eDe<j        jM        e<j        jN        g           e/dd          d8dtedeudeeef         fd                        Z eDe<j        jM        e<j        j        g           e/dddd          dtedeeeeef         fd                        Z eDe<j        jM                  	 	 	 d9dtedeYdeYde	eu         fd            ZdededeeWeX         eWeX         f         fdZdededue	eu         deeef         fdZd<ededeYfdZ eDe<j                  dRdddddddted|ed}eYdeYde	e         de	e         de	e         de	e         deeeeef         fd            Z eDe<j        jM        e<j        jN        g          dRdddÜdted|edeYd}eYdeYde	e         defdƄ            Z eDe<j                   e/dddRɦ          	 	 	 d:d(edtedeYdeYdeYdeeef         fd˄                        Z eDe<j        jM                  d̄             Z eDe<j                   e/            	 	 d;d<ededed}eYdeYdefd̈́                        Zd΄ Zdτ Z eDe<j                   e/            dЄ                         Z eDe<j                   e/            dф                         Zd҄ Z eDe<j                   e/dӦ          dԄ                         Z eDe<j                   e/dӦ          dՄ                         Zdք Z eDe<j                   e/            dׄ                         Z eDe<j                   e/            d؄                         Z eDe<j        jM        e<j        j        e<j        jM        e<j        j        g           e/dӦ          dل                         Zdڄ Z eDe<j                   e/            dۄ                         Z eDe<j                   e/            d܄                         Z eDe<j        jM        e<j        j        e<j        jM        e<j        j        g           e/dӦ          d݄                         Z eDe<j                   e/            d<d(ededefd                        Z eDe<j                   e/            ded(edededef
d                        Z eDe<j        jM        e<j        jN        g           e/dRɦ          dGdGdd                        Z eDe<j        jM        e<j        jN        g           e/            dd/d                        Z eDe<j        j                  d=d            Z eDe<j        j                  d=d            Z  eDe<j        jM        e<j        jN        g           e/            d/d                        Z eDe<j        jM                  	 	 d6d            Z eDe<j                   e/dRɦ          d                         Zd Zd>dZ	 d/dej        d=ej        deeWeX         eXf         deeWeX         eXf         deeWeX         eXf         deYdeXde	eeWeX         eXf                  fdZ	d Z
 eDe<j        jM                  dej        d=ej        d?e	ej                 de	ej                 de	ej                 deYdedefd            Z eDe<j        jM                  dej        d=ej        d?ej        deWeX         deWeX         deWeX         deYdeWeX         deXfd            Zej        j        rej=        >                    d dd          Z eDej;        j        j        jM                  d             Z eDej;        j        j        jM                  d             Zej        j        rEej=        >                    ddd          Z eDej;        j        j                  d             Zej=        >                    ddd          Z eDej;        j        j        jM                   eDej;        j        j        jM                  d                         Z  eDej;        j        j        j!                  d             Z" eDej;        j        j#        jM                   eDej;        j        j#        j$                  d                         Z% eDej;        j        j#        j!                   eDej;        j        j#        j&                  d	                         Z' eDej;        j        j(        jM                   eDej;        j        j)        jM                  d
                         Z*ej=        >                    ddd          Z+ eDej;        j,        j-                  	 	 	 	 d?d            Z. eDej;        j,        j/                  d             Z0d Z1 eDe<j2        jM                  	 	 	 	 	 d@d            Z3d Z4 eDe<j5        jM                  d             Z6 eDe<j7                   e/            	 	 	 	 	 d@d                        Z8 eDe<j9                   e/dӦ          d                         Z: eDe<j;        jM                  d             Z< eDe<j=        jM                  d             Z> eDe<j?        jM                  d             Z@ eDe<jA                   e/dӦ          d                         ZBdedzeufdZC eDe<jD                   e/dd$          d                         ZE eDe<jF                   e/dӦ          d                         ZG eDe<jH                   e/dd$          d                         ZI eDe<jJ                   e/dӦ          d                          ZK eDe<jL        j                  d/d!            ZM eDe<jN        jM        e<jN        jN        g           e/            d"                         ZO eDe<jP        jM        e<jP        jN        g           e/            dd#d$eXfd%                        ZP eDej;        j<        jQ        jM        ej;        j<        jQ        jN        g           e/            d&                         ZQ eDe<jR        j        e<jS        j        g          d'             ZT eDe<jU        jM        g          d(             ZV eDe<jW        jM        e<jW        jN        g           e/dRɦ          dGdGdd)                        ZX eDe<jY        j        g          d*             ZZ eDe<j[        jM        e<j\        jM        g          ddd+d,            Z] eDe<j^        jM        g          ddd+d-            Z_ eDe<j`        g           e/            d.                         Za eDe<jb        g          d/             Zc eDe<jd        g          d0             Ze eDe<jf        g          d1             Zg eDe<jh        g          d2             Zi eDe<jj        g          d3             Zjd4eXd5eXdeXfd6Zkd7 Zl eDe<jm        g          d?e	e         fd8            Zn eDe<jo        g          d9             Zp eDe<jq        g          d:             Zr eDe<js        jM                  d;             Zt eDe<ju                   e/            d<                         Zv eDe<jw        jM                  	 	 	 	 	 	 dAd=            Zx eDe<jy        jM                  d>             Zzd2d?Z{ eDe<j|        jM        e<j|        jN        g           e/            dBdd@dA                        Z} eDe<j~        jM        e<j        jM        g          dB             Z eDe<j~        j        e<j~        j        e<j        j        e<j        j        e<j        jM        e<j        j        g           e/d#d$          dCdC                        Z eDe<j        jM                  dD             Z eDe<j        jM                  dE             Z eDe<j        jM                  dF             Z eDe<j        j        e<j        j        e<j        j        e<j        j        e<j        jM        e<j        jM        e<j        jM        g          dG             Z eDe<j        j        e<j        j        e<j        j        e<j        j        g          ddH            Z eDe<j        j        e<j        j        g          ddI            Z eDe<j        jM        e<j        j        g          dJ             ZdK Z eDe<j        j        e<j        j        g          dL             Z eDe<j        j        e<j        j        g          dM             Z eDe<j        jM                  dN             Z eDe<j        j        e<j        j        g          dO             Z eDe<j        j        e<j        j        g          dP             Z eDe<j        jM                  dQ             Z eDe<j        j                   e/            ddefdR                        Z eDe<j        g           e/            	 dDdU                        Z eDe<j        g          	 dDdV            Z eDe<j        g          	 dDdW            Z eDe<j        jM        e<j        jM        g          d-dX            Z eDe<j        j                  dY             Z eDe<j        jM                  dZ             Z eDe<j                  d[             Z eDe<j                   e/            d\                         Z eDe<j                  d]             Z eDe<j        jM                  d-d^            Zd1d_Z eDe<j        jM                  d`             Z eDe<j        jv                  da             Zdb ZÐdc ZĐdd ZŐde Z	 d-d<edfeXdgeXdheXdieXdjeXdkeXdleXdmeXdneXdoeXdpeXdqeXdreXdseXdteXdueXdveXdweXdxeXdeudyeYf,dzZǐd{ Zd<ededfeXdgeXdheXdieXdjeXdkeXdleXdmeXdneXdoeXdseXdteXdueXdveXdweXdxeXdeuf&d|Zɐd} Z eDe<j        jM                  d~             Z eDe<j        jM                  	 	 	 	 d?d            Z eDe<j        jM                  d             Z eDe<j                   e/dd$          	 	 	 	 d?d                        Z eDe<j                   e/dӦ          d                         Zd<edefdZ G d de          Zd<ededeXfdZ eDe<j        jM                  d             Z eDe<j                   e/            d                         Z eDe<j                   e/dӐd          d                         Z eDe<j        jM        g          d             Z eDe<j        jM                  	 	 	 	 	 dEd            Z eDe<j        jM        e<j        jN        g           e/            ddddddd                        Z eDe<j        jM        e<j        jN        g           e/            ddddddd                        Z eDe<j        jM                  d             Z eDe<j        jM                  dFd            Zd2d)eXdeXdeYfdZd Zd Z eDe<j        jM                  d-d            Zd-dZd/dZd Zd/dZdGdZ eDe<j        jM                  d             Z eDe<j                  d             Z eDe<j        j        e<j        j        e<j        j        e<j        j        g           e/            d/d                        Z eDe<j        j        e<j        j        e<j        j        e<j        j        g          d/d            Z eDe<j        g          	 	 	 	 dHdededededeYdeYde	e         fd            ZdedeeXdf         fdZ eDe<j        g          	 	 	 	 dHdededede	e         deYdedeYdeYde	e         fd            Z  eDe<j        g          	 	 	 	 	 dIdededede	e         dedeYdeYde	e         fd            Z eDe<j        g          	 d/dededededededededeXdeXdedeYdedede	e         fd            Z eDe<j        g          	 	 	 	 dJdededededeYde	e         de	e         fd            Z eDe<j        g          	 	 d1dedededededededeYde	e         de	e         fd            Z eDe<j	        g          	 	 	 	 	 dKdededede	e         dedeYde	e         de	e         deeef         fd            Z
 eDe<j        g          	 	 	 dLdededede	e         deYdeYde	e         fd            Z eDe<j        g          	 	 dMdedededede	e         dedededededeWeY         deYde	e         fdÄ            Z eDe<j        g          	 d/dedededededededededededeXdeXdedeYde	e         f dĄ            Z eDe<j        g          	 	 	 	 	 dEdededede	e         de	e         deXdeXdedeYdeYde	e         de	eX         de	eX         de	e         de	e         fdɄ            Z eDe<j        g          	 	 	 d.dededededededededeXdeXdedeYdedede	e         de	eX         de	eX         f"dʄ            Z eDe<j        g          	 	 	 	 	 dNdededed?e	e         de	e         de	e         de	eX         de	eX         dedeXdeYde	e         de	e         de	e         de	eX         fdӄ            Z eDe<j        g          	 	 	 dGdedededed?e	e         de	e         de	e         dej        dej        dededededeXdeYde	e         de	eX         deYf$dׄ            Z eDe<j        jM        g          	 	 	 	 dOd(ej        dEej        dej        dej        d?e	ej                 de	ej                 dAe	ejv                 deYfd܄            Z eDe<j        j        e<j        j        g           e/            d2d݄                        Z eDe<j         j                  d2dބ            Z! eDe<j"        jM        e<j"        jN        g           e/            d-dd/d߄                        Z#d Z$d Z% eDe<j&        jM        e<j'        jM        g          d/d            Z& eDe<j(        jM        e<j)        jM        g          d1d            Z( eDe<j*        jM        e<j+        jM        g          	 	 d1dedeeeXej        f                  deeeXej        f                  de	e         de	e         f
d            Z* eDe<j,        jM        e<j-        jM        g          d.d            Z, eDe<j.        jM        e<j.        j/        e<j.        j        e<j.        j0        g          dPd            Z1d Z2 eDe<j3        jM                  	 	 d1d            Z4 eDe<j5        jM                  d             Z5 eDe<j6        jM                  d             Z6d Z7d Z8 eDe<j9        jM        e<j:        jM        g          dBd            Z; eDe<j<        jM                  dQd            Z< eDe<j=        jM                  dRd            Z> eDe<j?                   e/            	 dSd                        Z@ eDe<jA        jM        e<jA        j        g           e/d#d$          dCd                        ZBejC        ZDd ZE eDe<jF        jM                  d             ZF eDe<jG        jM                  d             ZG eDe<jH        jM                  d             ZI eDe<jJ        jM                  d             ZJ eDe<jK        j        e<jK        jL        g           e/            dddd                        ZM eDe<jN        g           e/            dTd                        ZO eDe<jP        jM        e<jQ        jM        g          	 	 d1d            ZR eDe<jS        jM        g          	 	 d1d             ZT eDe<jU        jM                  d             ZU eDe<jV        jM        e<jV        jN        g           e/            d.d                        ZV eDej;        j<        jW                  d             ZW eDej;        j<        jX                  d             ZX eDe<jY                   e/            dddddd                        ZZd Z[ eDe<j\                  d             Z] eDe<j^                  	 dUd	            Z_ eDe<j`                  	 dUd
            Za eDe<jb                  	 dUd            Zc eDe<jd                   e/            dddd                        Ze eDe<jf                   e/            deXd(edefd                        Zg eDe<jh                  d(efd            Zi eDe<jj                   e/dRɦ          d(edefd                        Zj eDe<jk                   e/            d(edefd                        Zkd Zl	 	 	 	 	 dVdedede	ej                 de	ej                 de	e         d?e	e         de	ej                 dAe	ejv                 deYfdZm eDe<jn                   e/            	 	 	 d.dedede	e         d?e	e         dAe	ejv                 defd                        Zo eDe<jp        g          	 	 	 	 	 dVdej        dej        dej        dej        de	ej                 d?e	ej                 de	ej                 dAe	ejv                 deYfd            Zq eDe<jr                   e/            ded)eXdeYdefd                        Zs eDe<jt                   e/            dd                        Zu eDe<jv                   e/            	 	 	 dWd=ed$edeXdeYd eYdefd!                        Zv eDe<jw        jM                  	 dXd#ed^eWe         d"eWeX         d#efd$            Zxd% Zyd& Zz eye<j{                    eye<j|                    eye<j}                    eye<j~                    eye<j                    eye<j                    eye<j                    eye<j                    eye<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                    eze<j                   d' Z eDe<j                   e/            d(                         Z eDe<j                   e/            dGd)d*                        Z eDe<j                   e/            dGd)d+                        Z ee<j                  Z ee<j                  Z ee<j                  Zd dl0Zd dlZd dlZd, Z e             dS (Y      N)Sequence)Enum)wraps)CallableOptionalTypeVarUnion)	ParamSpec)SymBoolSymFloatTensor)_add_op_to_registry_convert_out_paramsglobal_decomposition_table
meta_table)
OpOverload)_prim_elementwise_meta$ELEMENTWISE_PRIM_TYPE_PROMOTION_KIND)BoolLikecorresponding_complex_dtypecorresponding_real_dtypeelementwise_dtypesELEMENTWISE_TYPE_PROMOTION_KIND	FloatLikeIntLikemake_contiguous_strides_forNumbersuggest_memory_format
TensorLike)_maybe_convert_to_dtype_maybe_resize_out_resize_output_check_safe_copy_outout_wrapper)_broadcast_shapes_maybe_broadcast)_config)_pytree_T_PatenIMPLMeta   returnc                       fd}|S )Nc                 \     t                       fd}t          j        |            S )Nc                 4    t          t          |            d S N)r   r   )opfns    S/root/voice-cloning/.venv/lib/python3.11/site-packages/torch/_meta_registrations.pyregisterz0register_meta.<locals>.wrapper.<locals>.register:   s    
B33333    )r   pytree	tree_map_)r5   r7   r4   s   ` r6   wrapperzregister_meta.<locals>.wrapper7   sD     $$	4 	4 	4 	4 	4 	2&&&	r8    )r4   r;   s   ` r6   register_metar=   6   s#         Nr8   type_promotionc                     t          j        |d| i\  }fd|D             }t          | }t          |dt          j        iS )Ntype_promotion_kindc                 0    g | ]}t          |          S r<   )r    ).0xresult_dtypes     r6   
<listcomp>z$elementwise_meta.<locals>.<listcomp>L   s$    CCC#A|44CCCr8   r>   )utilsr   r&   r   r   DEFAULT)r>   args_rD   s      @r6   elementwise_metarJ   C   sp    
 .	* OA| DCCCdCCCD T"D "	BJ  r8   c                     t           j        t           j        t           j        t           j        t           j        t           j        i}|                    | |           S r3   )torch	complex32halfcfloatfloatcdoubledoubleget)dtypefrom_complexs     r6   toRealValueTyperV   W   s<    eku|L
 E5)))r8   c                 x     t          t           g|R            t          j         k     fd           d S )Nc                      d d  S )Nzoutput with shape z# doesn't match the broadcast shape r<   )broadcasted_shape
self_shapes   r6   <lambda>z)check_inplace_broadcast.<locals>.<lambda>d   s    gZggTegg r8   )tupler%   rL   _check)rZ   
args_shaperY   s   ` @r6   check_inplace_broadcastr_   `   sX    /
HZHHHII	LZ'ggggg    r8   Fc	                 \   	 t           t          j                  r,t          j                                         dk    d            t          t          j                  r,t          j                                        dk    d            t          d  fD                       rVt          j        t          j                              		nAt          j        t          j	                  	fd           npt          j                    t          t          j
                  sJ t          j        t          t                     fd           t          t                    sJ t          j        dk    d            t          j        f|d||	          S )
Nr   c                      dS Nz:linspace only supports 0-dimensional start and end tensorsr<   r<   r8   r6   r[   z(meta_linspace_logspace.<locals>.<lambda>x       P r8   c                      dS rb   r<   r<   r8   r6   r[   z(meta_linspace_logspace.<locals>.<lambda>}   rc   r8   c              3   @   K   | ]}t          |t                    V  d S r3   )
isinstancecomplex)rB   args     r6   	<genexpr>z)meta_linspace_logspace.<locals>.<genexpr>   s,      
C
C:c7##
C
C
C
C
C
Cr8   c                      d  d S )Nzlinspace(): inferred dtype z& can't be safely cast to passed dtype r<   )default_complex_dtyperT   s   r6   r[   z(meta_linspace_logspace.<locals>.<lambda>   s    z6Kzzsxzz r8   c                      dt                    j         dt                     j         dt                    j         dS )Nz4received an invalid combination of arguments - got (, ))type__name__)endstartstepss   r6   r[   z(meta_linspace_logspace.<locals>.<lambda>   sY     Hu++H Hs)),H H04U0DH H H r8   c                      dS )Nz$number of steps must be non-negativer<   r<   r8   r6   r[   z(meta_linspace_logspace.<locals>.<lambda>   s    %K r8   metarT   layoutdevice
pin_memoryrequires_grad)rf   rL   r   r]   dimanyrF   r   get_default_dtypeis_complex_dtyperT   _check_typer   empty)
rr   rq   rs   baserT   rx   rw   ry   rz   rk   s
   ``` `    @r6   meta_linspace_logspacer   h   s    %&& 
IIKK1PP	
 	
 	
 #u|$$ 
GGIINPP	
 	
 	

 
C
CsE/B
C
C
CCC 3 % A#%%!
 !
 =)EEL&u--zzzzz   
 2022eU[))))) 
5'""	H 	H 	H 	H 	H 	H  
 eW%%%%%	L!KKLLL;	#   r8   c                    t          j        j        t           j        k    fd           t          j        |                                 dk    o                                dk     d            |                     j                  S )Nc                      d j          S )Nz2take(): Expected a long tensor for index, but got rT   indexs   r6   r[   zmeta_take.<locals>.<lambda>   s    RU[RR r8   r   c                      dS )Nz*take(): tried to take from an empty tensorr<   r<   r8   r6   r[   zmeta_take.<locals>.<lambda>   s    < r8   )rL   r]   rT   long_check_indexnumel	new_emptyshape)selfr   s    `r6   	meta_taker      s     
Luz!RRRR  
 
ZZ\\Q55;;==A#56<<   >>%+&&&r8   r{   c                J     j         }j         }t          j        ||k    d            t          j                                       dk    o                              dk     fd           t	           j        j                  }                     |          S )Nc                      dS )Nz=linalg.cross: inputs must have the same number of dimensions.r<   r<   r8   r6   r[   zlinalg_cross.<locals>.<lambda>       O r8   r.   c                  f    d  d                                 d                                 S )Nzlinalg.cross: inputs dimension z must have length 3. Got  and size)r{   otherr   s   r6   r[   zlinalg_cross.<locals>.<lambda>   sG    :c : :99S>>: :(-

3: : r8   )ndimrL   r]   r   r%   r   r   )r   r   r{   x_dy_d	out_shapes   ```   r6   linalg_crossr      s     )C
*C	Ls
OO   
L		#!4

31 4	
 	
 	
 	
 	
 	
   "$*ek::I>>)$$$r8   c                     t          | d           t          | d           t          j        | t          j                  S )Nzlinalg.matrix_expmemory_format)squareCheckInputscheckFloatingOrComplexrL   
empty_likecontiguous_formatr   s    r6   linalg_matrix_expr      s?     d/0004!4555D0GHHHHr8   valuesindicesc                    t          j        | j        | j        | j                  }t          j        | j        | j        t           j                  }|                                 dk    r | j        dk    rt          || j                   ||fS )Nrx   rT   r   )	rL   r   r   rx   rT   int64r   r   maybe_wrap_dim)r   r{   r   r   s       r6   	cummaxminr      su    
 [DKtzJJJFk$*T[LLLGzz||qTY!^^sDI&&&7?r8   c                 l    t          || j                   t          j        | t          j                  S Nr   )r   r   rL   r   r   )r   r{   s     r6   logcumsumexpr      s/     3	"""D0GHHHHr8   c                   |j         }t          |          }||z
  }t          t          |                    }d t          |          D             }	|D ]}
d|	|
<   g g }}|D ]5}
|	|
         s|                    |
            |                    |
           6||z   }t          |          }|                                |d |         }|                    fdd           |||d          z   }|                    |          }dgt          |j        |d                    z   }|	                    |          }|
                    d          }||d<   t          |          }t          t          |                    D ]}|||                  ||dz   <   |                     |t          j                   d	 t          |          D             }d}|dz
  }|dk    r=||                     d          z  |||         <   ||||                  z  }|dz  }|dk    =t          ||          D ]&}|                     d||z
  z             |||         <   '|                     |||                                            | S )
Nc                     g | ]}d S Fr<   rB   rI   s     r6   rE   z_exec_fft.<locals>.<listcomp>   s    555A%555r8   Tc                     |          S r3   r<   )rC   self_stridess    r6   r[   z_exec_fft.<locals>.<lambda>   s    <? r8   keyreverser   r      r   c                     g | ]}d S r   r<   r   s     r6   rE   z_exec_fft.<locals>.<listcomp>  s    ***1***r8   )r   lenlistrangeappendstridesortpermuter   reshaper   resize_rL   r   as_strided_storage_offset)outr   	out_sizesr{   forwardr   signal_ndim
batch_dimsdim_permuteis_transformed_dimdleftright	batch_endtmpinputbatched_sizes
batch_sizebatched_out_sizesiout_stridesbatch_numelr   s                         @r6   	_exec_fftr      s   9Dc((K#J uT{{##K55t555 % % $1 b%D  !!$ 	KKNNNNLLOOOO,KD		I;;==L
jyj
!CHH****DH999IJJ//KLL%%E D4JKK 8999MMM-((EAJ!M!]++3s88__ 5 5#,SV#4!a%  KK!1HKIII +*eDkk***KKQA
q&&&1CJJqMM&AKN#yQ00	Q q&& :t$$ G G&)jja*n1E&F&FKN##OOI{C,>,>,@,@AAAJr8   r   r{   exclude_lastc                     t          |          }|                                 |d t          |          t          |          z
                               fd           |S )Nc                     |          S r3   r<   )r   r   s    r6   r[   z_sort_dims.<locals>.<lambda>"  s    l1o r8   )r   )r   r   r   intr   )r   r{   r   sorted_dimsr   s       @r6   
_sort_dimsr     sl    s))K;;==L6#k""S%6%6667<<%%%% =    r8   c                 $   t          j        | j        j                   |s|                                 S t          | |          }|                     |                                           }t          || |                                 ||          S )Nr   )	rL   r]   rT   
is_complexcloner   r   r   r   )r   r{   normalizationr   r   r   s         r6   meta_fft_c2cr   )  sx     
L&''' zz||T3''K
..
%
%CS$		['JJJJr8   c                     t          |           t          k    s+t          |           dk    r| d         dk    r| d         dk    rdS dS )N   r   r   FT)r   cufft_max_ndimr   s    r6   use_optimized_cufft_pathr   8  sD    
3xx.  SXX]]s1v{{s1vQR{{utr8   c                 6   t          j        | j        j                   t	          |                                           }t	          |          }|d         }||         dz  dz   }t	          |          }|||<   |r|||<   t          |           dk    st          |           dk    r|                     |t          j	        | j                            }	| }
t          |           dk    r%t          |          rt          |	|
||d           nt          |          dk    r|n|}t          |	|
||gd           t          |          dk    r.|                     |t          j	        | j                            }
|d d         }|r|
|	}
}	|
                                |                    fd	d
           t          t           t          |                    }|t          |          |z
  d          }t          |	|
||d           |d t          |          |z
           }||sB|	                    |          ||         k    r#|
                    |t           j                   |
}	|	S |                     |t          j	        | j                            S )Nr   r   r   cudaxpur   Tr   c                     |          S r3   r<   )r   stridess    r6   r[   zmeta_fft_r2c.<locals>.<lambda>f  s    '!* r8   r   r   )rL   r]   rT   is_floating_pointr   r   device_hintr   rF   r   r   r   r   r   r   minr   r   r   )r   r{   r   onesidedinput_sizesr   last_dimlast_dim_halfsizeonesided_sizesoutputworking_tensortarget_sizesr   max_dims	last_dimsr   s                  @r6   meta_fft_r2cr   ?  s    
L-...tyy{{##K[!!I2wH#H-2Q6+&&N0N8 0/	(4F""k$&7&75&@&@ U>tzJJ   
 
 t&&+CC+H+H&fnidKKKKK ),CA99>LfnlXJPTUUUU3xx!||!%U%Ftz%R%R "0 " "
 crc(K I)7(//11  ,,,,d !    ~s;/?/?@@'K(8(88(C(E(EF	NNIt    **GC,<,<x,G*GH  I  	({{8$$	((;;;&&y@W&XXX' ~~U>tzJJ  
 
 	
r8   )	generatorc                H    t          |t          j        | g                    S r3   )r!   rL   Size)nr   r   s      r6   meta_randpermr  |  s    S%*aS//222r8   rT   rw   rx   ry   c                4    t          j        | ||||          S Nr  rL   r   )r  rT   rw   rx   ry   s        r6   meta_randperm_defaultr	    s(     ;	vf   r8   c                v     dt          j         k     fd           t          j        |||||          S )Nr   c                      d d  S Nz:random_ expects 'from' to be less than 'to', but got from=z >= to=r<   highlows   r6   r[   zmeta_randint.<locals>.<lambda>      _S__Y]__ r8   r  rL   r]   r   )r  r   rT   rw   rx   ry   r  s   `     @r6   meta_randintr    s[     C	Ls
_____   ;E&J   r8   c                r     t          j         k     fd           t          j        |||||          S )Nc                      d d  S r  r<   r  s   r6   r[   z"meta_randint_low.<locals>.<lambda>  r  r8   r  r  )r  r  r   rT   rw   rx   ry   s   ``     r6   meta_randint_lowr    sV     
Ls
_____   ;E&J   r8   c                4    t          j        | ||||          S r  r  )r   rT   rw   rx   ry   s        r6   meta_rand_defaultr    s(     ;E&J   r8   r   lastdimc                    t          j        | j        j                   t	          |           dk    rt          |                                           }|||d         <   |                     |t          | j                            }t          |          r2t          ||                     t           j                  ||d          S t          |          dk    rt          | |d d         d|          }n |                     t           j                  }t          ||||d         gd          S | }t          |          dk    r'|d d         }t          | ||d          }|dd          }t          |                                          }|||d         <   |                     |t          | j                            }	t          |	|||d          S )	Nr   r   r   r   Fr   r   r   )rL   r]   rT   r   r   r   r   r   rV   r   r   r   r   r   r   )
r   r{   r   r  r   r   tempr   c2c_dimsr   s
             r6   meta_fft_c2rr    s    
L&'''4F""%%	$	#b'	1L1LMM#C(( 	P

)@
AA    3xx!||#D#crc(Aw??zz0GzHHVT9s2wiOOOO s88a<<3B3xH xNNNEbcc(C&&	$	#b'nnYodj.I.InJJeYUCCCCr8   c                    ddl m}  ||           s't          j        |           dk    rt	          d          t          |t                    rr|                    | |          }|                                 |                                k    r2t          j
                            ||                                            | S )Nr   )free_unbacked_symbolsr   zQmore than one element of the written-to tensor refers to a single memory location)%torch.fx.experimental.symbolic_shapesr  rL   _debug_has_internal_overlapRuntimeErrorrf   r   tor   r+   expand_copydefault)r   srcnon_blockingr  intermediates        r6   
meta_copy_r(    s     LKKKKK "!$''
,1,Md,S,SWX,X,X_
 
 	
 #v @vvdL1199;;,++----$$\499;;???Kr8   c                 <   t          |                                           }t          |                                           }||                                 k    rdn||         ||         z  }|                    |d           |                    ||           ||fS Nr   )r   r   r   r{   insert)tensorr{   result_sizesresult_strides
new_strides        r6   inferUnsqueezeGeometryr0    s    &&L&--//**NVZZ\\))|C/@>RUCV/VJQ#z***''r8   c                     t          ||                                 dz             }t          | |          \  }}|                     ||           | S r*  )r   r{   r0  r   )r   r{   g_sizes	g_stridess       r6   meta_unsqueeze_r4    sL    
dhhjj1n
-
-C/c::GYWi(((Kr8   r   weight_metabias_activation_opt	out_dtypec                 l   t          | j                  }|6|                    d          |                    d          k    s
J d            |                    d          |                     d          dz  k    sJ |                    d          |d<   t          | j                  dk    s
J d            d|                     d          f}|/| j        t
          j        k    r|t
          j        k    s
J d            |                     ||| j        n|          	                    ||          }|S )	Nr   zoutput size mismatchr   r   r   z*we can only handle the squashed input case9out_dtype is only supported for i8i8->i32 linear operatorr   )
r   r   r   r   rT   rL   int8int32r   
as_strided)	r   r5  r6  r7  r8  r9  output_sizestransposed_stridesr   s	            r6   meta_sparse_structured_linearrA    s:    $$L{{1~~1---/E---;;q>>UZZ^^a/////{{1~~L u{q   "N   UZZ]]+{ej((Y%+-E-E-EG .F-EE __&.ekkI    j122 
 Mr8   mat1	mat1_metamat2c                    t          | j                  dk    sJ t          |j                  dk    sJ t          |j                  dk    sJ |                     d          |                    d          dz  k    sJ |                     d          |                    d          g}|/|j        t          j        k    r|t          j        k    s
J d            |                    |||j        n|          }|S )Nr   r   r   r;  r   r   r   r   rT   rL   r<  r=  r   )rB  rC  rD  r9  r?  r   s         r6   meta_sparse_structured_mmrG  6  s     tz??ay1$$$$tz??a99Q<<499Q<<!+++++IIaLL$))A,,/LzUZ''I,D,D,DG -E,DD ^^%-djj9   F
 Mr8   r   )alphabetar9  c                   t          | j                  dk    s
J d            t          |j                  dk    sJ t          |j                  dk    sJ t          |j                  dk    sJ |                     d          |                    d          k    s
J d            |                    d          |                    d          dz  k    sJ |                    d          |                    d          g}|/|j        t          j        k    r|t          j        k    s
J d            |                    |||j        n|          }|S )Nr   zEonly input broadcasted to columns of mat1 * mat2 product is supportedr   r   r;  r   rF  )	r   rB  rC  rD  rH  rI  r9  r?  r   s	            r6   meta_sparse_structured_addmmrK  O  sX    u{q   O !   tz??ay1$$$$tz??a::a==DIIaLL(((O )(( 99Q<<499Q<<!+++++IIaLL$))A,,/LzUZ''I,D,D,DG -E,DD ^^%-djj9   F
 Mr8   compressed_Adense_BrH  transpose_resultalg_idsplit_ksplit_k_modec	                    |j         t          j        t          j        t          j        t          j        t          j        hv s
J d            | j         |j         k    s
J d            t          |j                  dk    s
J d            | j         t          j        t          j        fv }	|	rdnd}
|	r|	                                r
J d            |
                    d          }|
                    d	          }|                                 d
z  |
|z  z  }|||
                    d          k    sJ |<|	r0|t          j        t          j        t          j        t          j        hv s
J d            |r||fn||f}|                    ||          S )Nz;_cslt_sparse_mm only supports fp16, bf16, int8, and fp8e4m3zinputs must have the same dtyper   z'_cslt_sparse_mm only supports 2d inputs
   	   z.dense input must be transposed for 8bit dtypesr   r      z\out_dtype is not supported for {compressed_A.dtype} x {dense_B.dtype} -> {out_dtype} matmul!r   )rT   rL   float32float16bfloat16r<  float8_e4m3fnr   r   is_contiguousr   r   r=  r   )rL  rM  r7  rH  r9  rN  rO  rP  rQ  is_8bit_input_typecompression_factorkr  moutput_shapes                  r6   meta__cslt_sparse_mmr`  r  s    =
    E   ...0Q...w}"""$M"""%+
E<O/PP18q 
((** 	
 	
<	
 	
* 	QAQA						"(:Q(>?ADIIaLL    ! 	
iMNK	4
 '
 '
 '
 k'
 '
 
 .9Aq66Aq6L\;;;r8   T)include_selfr   sourcereducera  c                B    t          j        | t           j                  S r   rL   r   r   r   r{   r   rb  rc  ra  s         r6   meta_index_reducerg    s     D0GHHHHr8   c                    | S r3   r<   rf  s         r6   meta_index_reduce_ri    s	     Kr8   c                     t          |                                           }|                                 dk    r|                                ||<   |                     |          S Nr   )r   r   r{   r   r   )r   r{   r   result_sizes       r6   meta_index_selectrm    sN     tyy{{##KxxzzA~~ ;;==C>>+&&&r8   )lengthsr   offsetsaxisunsafeinitialdatarn  ro  rp  rq  c                     |t          d           fd}| ||j                  S |,|j        d d         |j        d         dz
  fz   }	 ||	          S t          d          )Nz?segment_reduce(): indices based reduction is not supported yet.c                 x    t          j        | j        dz   d          z   j        dt           j                  S )Nr   ru   rT   rx   r   )rL   r   r   rT   r   )lengths_shaperp  rs  s    r6   segment_reduce_lengths_tensorz:meta_segment_reduce.<locals>.segment_reduce_lengths_tensor  sA    {DJtaxzz22*1	
 
 
 	
r8   r   r   z<segment_reduce(): Either lengths or offsets must be defined.)NotImplementedErrorr   r!  )
rs  rc  rn  r   ro  rp  rq  rr  rx  rw  s
   `    `    r6   meta_segment_reducerz    s     !M
 
 	

 
 
 
 
 
 ,,W];;; crc*gmB.?!.C-EE,,];;;
U
V
VVr8   c                 ,    |                      d          S Nr<   r   r   s    r6   meta_maxr~         >>"r8   c                     t          j        | j        |f          }t          | ||          }|                     |          |                     |t
          j                  fS Nr   rF   reduction_dimsr   _compute_reduction_shaper   rL   r   r   r{   keepdimr_  s       r6   meta_max_dimr    W    

tzC6
2
2C+D#w??L|$$|5:66 r8   c                 ,    |                      d          S r|  r}  r   s    r6   meta_minr    r  r8   c                     t          j        | j        |f          }t          | ||          }|                     |          |                     |t
          j                  fS r  r  r  s       r6   meta_min_dimr    r  r8   c                     |                                  rt          | j                  }nt          | t          j                  \  }}t          j        | |          S Nr@   r   )r   r   rT   r   r   INT_TO_FLOATrL   r   )r   rD   rI   s      r6   
meta_angler    s^     
/
;;, ? L
 
 
< D5555r8   c                     t          j        ||                                 | j                   |                    t          j        |                     S r3   )rL   _resize_output_r   rx   copy_angle)r   r   s     r6   meta_angle_outr    s=    	#tyy{{DK88899U[&&'''r8   c                     d S r3   r<   )vals    r6   assert_asyncr        
Fr8   c                     d S r3   r<   )r  
assert_msgs     r6   assert_async_metar  "  r  r8   c                     d S r3   r<   )ss    r6   
print_metar  '  r  r8   rT   rw   rx   ry   r   c                 .    t          j        dd          S )Nr   ru   rx   r  r  s        r6   make_dep_tokenr  ,  s     ;q((((r8   c                     ddl m} t          | t          t          f          rt          d           || ||           d S )Nr   )constrain_range'Constraining SymFloat or Symbool is nyir   max)r  r  rf   r   r   
ValueError)r   r   r  r  s       r6   sym_constrain_ranger  8  s[     FEEEEE$7+,, DBCCCODcs++++++r8   c                 @    t                               | ||           |S Nr  )r+   r  r   r   r  	dep_tokens       r6   functional_sym_constrain_ranger  B  s#    Ts444r8   c                 P   ddl m} ||t          j        |            d S t	          | t
          t          f          rt          d          t          |           t          u r6|t          j
        | |k               |t          j
        | |k               d S  || ||           d S )Nr   )_constrain_range_for_sizer  r  )r  r  rL   _check_is_sizerf   r   r   r  ro   r   r]   )r   r   r  r  s       r6   sym_constrain_range_for_sizer  H  s     POOOOO
{s{T"""$7+,, DBCCCDzzS?L%%%?L%%%d555555r8   c                 @    t                               | ||           |S r  )r+   r  r  s       r6   'functional_sym_constrain_range_for_sizer  \  s#    %%d%===r8   c                     |S r3   r<   )r  r  r  s      r6   functional_assert_async_metar  b  s    r8   f_namec                    |                                  dk    sJ | d            |                     d          |                     d          k    s9J | d|                     d           d|                     d           d            d S )Nr   z3: The input tensor must have at least 2 dimensions.r   z5: A must be batches of square matrices, but they are  by 	 matrices)r{   r   )r   r  s     r6   r   r   h  s    88::???FFF ?? 99R==DIIbMM)))ss		RTss[_[d[deg[h[hsss *))))r8   Anamec                     t          j         j        j        k     fd           t          j         j        j        k     fd           t          j                            d                              d          k    fd           t          j                            d                               d          k     fd           d S )Nc                  *    dj          d j          dS )Nz:Expected b and A to be on the same device, but found b on z
 and A on 	 instead.r  r  r   s   r6   r[   z(linearSolveCheckInputs.<locals>.<lambda>w  s0    :{: :&'h: : : r8   c                  *    dj          d j          dS )Nz=Expected b and A to have the same dtype, but found b of type z and A of type r  r   r  s   r6   r[   z(linearSolveCheckInputs.<locals>.<lambda>  s0    =z= =*+'= = = r8   r   r  c                  b    d                      d           d                      d           dS )Nz3A must be batches of square matrices, but they are r  r  r   r  r   r  s   r6   r[   z(linearSolveCheckInputs.<locals>.<lambda>  sF    BFF2JJB B,-FF2JJB B B r8   c                      d d                      d           d                      d           d                     d           d                     d           
S )NzIncompatible matrix sizes for z: each A matrix is r   r  z but each b matrix is r  r   )r  r  r   s   r6   r[   z(linearSolveCheckInputs.<locals>.<lambda>  s    HT H HH H)*H H%)YYr]]H H8<		"H H r8   )rL   r]   rx   rT   r   )r   r  r  s   ```r6   linearSolveCheckInputsr  t  s   	Lqx	
 	
 	
 	
 	
   
L
ag	
 	
 	
 	
 	
   
L	r

affRjj 	
 	
 	
 	
   
L	r

diimm#	
 	
 	
 	
 	
 	
    r8   tallow_low_precision_dtypesc                 ,   | j         t          j        |                                 p|                                 fd           |sIt          j        t          j        t          j        t          j        t          j        fv fd           d S d S )Nc                       d  S )Nz<: Expected a floating point or complex tensor as input. Got r<   rT   r  s   r6   r[   z(checkFloatingOrComplex.<locals>.<lambda>  s    6^^W\^^ r8   c                       d  S )Nz*: Low precision dtypes not supported. Got r<   r  s   r6   r[   z(checkFloatingOrComplex.<locals>.<lambda>  s    vPPPP r8   )	rT   rL   r]   r   r   rP   rR   rO   rQ   )r  r  r  rT   s    ` @r6   r   r     s    
 GE	L	/^^^^^   & 
ek5<u}MMPPPPP	
 	
 	
 	
 	

 
r8   arg_namec                 h    t          j        |                                 dk    fd           d S )Nr   c                       d  dS )Nz: The input tensor z! must have at least 2 dimensions.r<   )r  r  s   r6   r[   zcheckIsMatrix.<locals>.<lambda>  s    6YYhYYY r8   )rL   r]   r{   )r  r  r  s    ``r6   checkIsMatrixr    s@    	L	1YYYYY    r8   Br   c                 4    t                      t                     t          j        r,                     d                              d          k    n+                     d                              d          k     fd           d S )Nr  r   c                       drdnd d                      d           d                      d           d                     d           d                     d           d	S )
Nz2: Incompatible shapes of A and B for the equation zAX = BzXA = Bz (r  rC   r   r   rn   r   )r  r  r  r   s   r6   r[   z#checkInputsSolver.<locals>.<lambda>  s     J J-xxXJ JJ JffRjjJ J/0vvbzzJ J<=FF2JJJ J J r8   )r   r  rL   r]   r   )r  r  r   r  s   ````r6   checkInputsSolverr    s    a   !V	L$(Fr

affRjj  affRjjAFF2JJ.F	
 	
 	
 	
 	
 	
 	
    r8   resultfn_nameresult_namec                 `     t          j        j        j        k     fd           d S )Nc            	      8      d d dj          dj          	S )Nz: Expected z5 and input tensors to be on the same device, but got z on z and input on r  )r  r   r  r  s   r6   r[   z!checkSameDevice.<locals>.<lambda>  sV     L L; L LL L &L L=B\L L r8   )rL   r]   rx   )r  r  r   r  s   ````r6   checkSameDevicer    sU     
L%	
 	
 	
 	
 	
 	
 	
    r8   UPLOc                                                        }t          j        t                     dk    o|dk    p|dk     fd           d S )Nr   ULc                      d  S )Nz1Expected UPLO argument to be 'L' or 'U', but got r<   )r  s   r6   r[   zcheckUplo.<locals>.<lambda>  s    JDJJ r8   )upperrL   r]   r   )r  UPLO_uppercases   ` r6   	checkUplor    s[    ZZ\\N	LD		QKNc1J^s5JJJJJ    r8   eigenvalueseigenvectorsr  	compute_vc                    t          | d           t          |           t          | j                  }|r;|                     |          }|                    |t          |d                     n|                     dg          }|                                 |                     |t          | j	                            }||fS )Nzlinalg.eighF	row_majorr   r   )
r   r  r   r   r   r   r   poprV   rT   )r  r  r  r   vecsvalss         r6   meta__linalg_eighr    s     a'''dOOOMME  {{5!! ;EU S S STTTT{{A3	IIKKK;;uOAG$<$<;==D:r8   c                     t          | d           t          j        | j                  r| j        nt          j        | j                  }|                     | j        d d         |          S )Nzlinalg.eigvalsr   r   r   rF   r~   rT   r   r   r   )r   complex_dtypes     r6   meta__linalg_eigvalsr    sj     e-... !%+..	<.u{;; 
 ??5;ss+=?AAAr8   c                    t          | d           t          j        | j                  r| j        nt          j        | j                  }|                     | j        d d         |          }|                     | j        |          }||fS )Nz
linalg.eigr   r   r  )r   r  r   vectorss       r6   meta_linalg_eigr    s     e\*** !%+..	<.u{;; 
 __U["-]_CCFooeko??G7?r8   r%  c                 t    | j                             t          j                                      dd          S )Nr   r  r   )mTr   rL   r   	transpose)r%  s    r6   cloneBatchedColumnMajorr    s+    6<<e&=<>>HHRPPPr8   r  c                      t          |           S r3   )r  )r   r  r  s      r6   _cholesky_solve_helperr    s     #4(((r8   c                      t          j         j        dk     fd           t          j        j        dk    fd           t           d          \  }}t	          |||          S )Nr   c                      d j          dS )Nz-b should have at least 2 dimensions, but has  dimensions insteadr   r   s   r6   r[   z cholesky_solve.<locals>.<lambda>  s    ^	^^^ r8   c                      d j          dS )Nz-u should have at least 2 dimensions, but has r  r  r  s   r6   r[   z cholesky_solve.<locals>.<lambda>  s    [[[[ r8   cholesky_solve)rL   r]   r   !_linalg_broadcast_batch_dims_namer  )r   r  r  self_broadcastedA_broadcasteds   ``   r6   r  r    s     
L	Q^^^^   
L	![[[[   'Ha!' '#m ""2M5IIIr8   c                     |                                  dk    r t          j        | t          j                  S t	          | d           t          |           S )Nr   r   cholesky)r   rL   r   legacy_contiguous_formatr   r  r   r  s     r6   r  r    sN     zz||qE4RSSSSdJ'''"4(((r8   c                 @    t          | d           t          |           S )Ncholesky_inverse)r   r  r  s     r6   r
  r
  &  s#     d.///"4(((r8   check_errorsc                 D   t          | d           t          | d           | j        }t          |          }t	          |d          }|                     |          }|                    ||           |                     |d|dz
           t          j                  }||fS )Nzlinalg.choleskyFr   r   r   )	r   r   r   r   r   r   r   rL   r=  )r  r  r  A_shaper   	L_stridesr  infoss           r6   linalg_cholesky_exr  .  s    a*+++1/000gGw<<D ,GU;;I	GAMM'9%%% KKD1H-U[KAAEe8Or8   tauc                     t          j         j        dk    d            t          j                             d                               d          k    d            t          j                             d                              d          k    d            t          j         j        j        z
  dk     fd            j        dk    r: j        d d         }j        d d         t          j        |k    fd	           t          j        j         j        k     fd
           t          d d           t          j         j        t           j        d           j         j	                  S )Nr   c                      dS )NzHtorch.linalg.householder_product: input must have at least 2 dimensions.r<   r<   r8   r6   r[   z,linalg_householder_product.<locals>.<lambda>G      Z r8   r  r   c                      dS )Nzbtorch.linalg.householder_product: input.shape[-2] must be greater than or equal to input.shape[-1]r<   r<   r8   r6   r[   z,linalg_householder_product.<locals>.<lambda>K  s    t r8   c                      dS )Nz`torch.linalg.householder_product: input.shape[-1] must be greater than or equal to tau.shape[-1]r<   r<   r8   r6   r[   z,linalg_householder_product.<locals>.<lambda>O  s    r r8   r   c                  (    dj          d j          S )Nzptorch.linalg.householder_product: Expected tau to have one dimension less than input, but got tau.ndim equal to  and input.ndim is equal to r  r   r  s   r6   r[   z,linalg_householder_product.<locals>.<lambda>T  0    \),\ \OTz\ \ r8   c                      d  S )Nzltorch.linalg.householder_product: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r<   actual_batch_tau_shapes   r6   r[   z,linalg_householder_product.<locals>.<lambda>^      O6LO O r8   c                  (    dj          d j          S )Nz,torch.linalg.householder_product: tau dtype z does not match input dtype r   r  s   r6   r[   z,linalg_householder_product.<locals>.<lambda>f  s*    939 9 9+0;9 9 r8   z torch.linalg.householder_productr  Fr  r   r   rT   rx   )
rL   r]   r   r   r   rT   r  empty_stridedr   rx   )r   r  expected_batch_tau_shaper  s   `` @r6   linalg_householder_productr#  @  s   
 
L
aZZ   
L

2%**R..(tt   
L

2#((2,,&rr  
 
L
SX"	
 	
 	
 	
 	
   zA~~#(;ss#3 !$3B3"&>>   	
 	
 	
 
L	U[ 	
 	
 	
 	
 	
   6UEJJJ[*5;%HHHk|	   r8   c                 <   t          | d           t          | dd           |                     | j                  }|                    | j        t          | j        d                     |                     | j        d d         t          j                  }||fS )Nzlinalg.inv_exF)r  r  r  r   r   r   r   r   r   r   rL   r=  )r  r  r  r  s       r6   linalg_inv_ex_metar&  v  s    a)))1o%PPPP	AGAMM!'6qw%PPPQQQKKEKK88Ee8Or8   LDpivotsinfo)	hermitianr  r*  c                z   t          | d           t          | d           t          j        | j        t          | j        d          | j        | j                  }|                     | j        d d         t          j	                  }|                     | j        d d         t          j	                  }|||fS )Nztorch.linalg.ldl_factor_exFr  r   r   r   r  )
r   r   rL   r!  r   r   rT   rx   r   r   )r   r*  r  r'  r(  r)  s         r6   linalg_ldl_factor_ex_metar,    s     d89994!=>>>		Z*4:GGGj{	
 
 
B ^^DJssO59^==F>>$*SbS/>;;Dvtr8   )r*  c                @    t           d           t           d           t           d           t          j        j        dk    fd            j        d d         }t          j        |j        k    fd           t          j        t          j        j	                  fd           t          j         j	        j	        k     fd           t                     \  }}t          j        |t          |d	          j	        j        
          S )Nztorch.linalg.ldl_solver   c                      d j          dS )NzMtorch.linalg.ldl_solve: Expected B to have at least 2 dimensions, but it has r  r  )r  s   r6   r[   z'linalg_ldl_solve_meta.<locals>.<lambda>  !    6&6 6 6 r8   r   c                      d j          dS )Nzjtorch.linalg.ldl_solve: Expected LD.shape[:-1] and pivots.shape to be the same, but got pivots with shape  insteadr   r(  s   r6   r[   z'linalg_ldl_solve_meta.<locals>.<lambda>  %    @)/@ @ @ r8   c                      d j          S )Nz<torch.linalg.ldl_solve: Expected pivots to be integers. Got r   r3  s   r6   r[   z'linalg_ldl_solve_meta.<locals>.<lambda>  s    ]v|]] r8   c                  (    dj          d j          S )Nz!torch.linalg.ldl_solve: LD dtype z does not match b dtype r   )r  r'  s   r6   r[   z'linalg_ldl_solve_meta.<locals>.<lambda>  s    _BH__VWV]__ r8   Fr  r   )r   r   r  rL   r]   r   r   rF   is_integer_dtyperT   _linalg_broadcast_batch_dimsr!  r   rx   )r'  r(  r  r*  expected_pivots_shapeB_broadcast_sizerI   s   ```    r6   linalg_ldl_solve_metar;    sc    b2333278881b":;;;	L	!	
 	
 	
 	
   HSbSM	L-	
 	
 	
 	
   
Lv|,,]]]]   
L
AG_____   7q"==a*+;uMMMgx	   r8   Pr  )pivotr=  c                    t          j         j        dk     fd           t           j                  }|d         }|d         }t          ||          }||d<   |r                     |          }n                     dg          }||d<                        |          }||d<   ||d<                        |          }|||fS )Nr   c                      d j          dS )Nz@linalg.lu: Expected tensor with 2 or more dimensions. Got size: r1  r2  r  s   r6   r[   z linalg_lu_meta.<locals>.<lambda>  s    dSTSZddd r8   r  r   r   )rL   r]   r   r   r   r   r   )	r  r=  sizesr^  r  r]  r<  r  r  s	   `        r6   linalg_lu_metarA    s     
L	!dddd  
 MMEb	Ab	AAq		AE"I KKKKE"I	EAE"IE"I	EAa7Nr8   LU)r=  r  c                    t          j         j        dk     fd           t           j                  }|d         }|d         }t          j        |t          |d           j         j                  }|	                                 t          ||          |d<                        |t           j                  }|	                                                      |t           j                  }|||fS )	Nr   c                      d j          dS )NzFtorch.lu_factor: Expected tensor with 2 or more dimensions. Got size: r1  r2  r  s   r6   r[   z*linalg_lu_factor_ex_meta.<locals>.<lambda>  s    jYZY`jjj r8   r  r   Fr  r   r   )rL   r]   r   r   r   r!  r   rT   rx   r  r   r   r   )	r  r=  r  r@  r^  r  rB  r(  r)  s	   `        r6   linalg_lu_factor_ex_metarE    s     
L	!jjjj  
 MMEb	Ab	A		*5EBBBgx	
 
 
B 
IIKKKAq		E"I[[ei[00F 
IIKKK;;uEI;..Dvtr8   )r   adjointrF  c                    t           d           t          j         j        j        k     fd           t          j        j        t          j        k    d            t           d           t           |d           t          j                             d                              d          k    d            t          j         j        d d         j        k    fd           t                     \  }}t          j
        |t          ||           j        j        	          }|                                d
k    r*|s(|                                r|                                }|S )Nztorch.linalg.lu_solvec                  *    dj          d j          dS )NzPlinalg.lu_solve: Expected LU and B to have the same dtype, but found LU of type  and B of type r1  r   )r  rB  s   r6   r[   z&linalg_lu_solve_meta.<locals>.<lambda>  s6    O$&HO O=>WO O O r8   c                      dS )NzElinalg.lu_solve: pivots should be a Tensor of scalar type torch.int32r<   r<   r8   r6   r[   z&linalg_lu_solve_meta.<locals>.<lambda>  s    W r8   zlinalg.lu_solver   c                      dS )NzYlinalg.lu_solve: Number of pivots per batch should be same as the dimension of the matrixr<   r<   r8   r6   r[   z&linalg_lu_solve_meta.<locals>.<lambda>   s    k r8   c                      d j          dS )Nzclinalg.lu_solve: Expected LU.shape[:-1] and pivots.shape to be the same, but got pivots with shape r1  r2  r3  s   r6   r[   z&linalg_lu_solve_meta.<locals>.<lambda>&  r4  r8   r  r   r   )r   rL   r]   rT   r   r   r  r   r   r8  r!  r   rx   r   r   conj)rB  r(  r  r   rF  r:  rI   r  s   ```     r6   linalg_lu_solve_metarN    s    26777	L
AG	
 	
 	
 	
 	
   
L	!WW   b1222b!T#4555	L
v{{2&kk   
L
"%	
 	
 	
 	
   7q"==a *+;4xPPPgx	  F ||~~4 	#[[]]FMr8   unpack_dataunpack_pivotsc                 D    t          j         j        dk     fd           |r)t          j        |j        t           j        k    d            t           j                  }|d         }|d         }t          ||          }||d<   |r                     |          }n                     dg          }|r:||d<                        |          }	||d<   ||d<                        |          }
n,                     dg          }	                     dg          }
||	|
fS )Nr   c                      d j          dS )NzFtorch.lu_unpack: Expected tensor with 2 or more dimensions. Got size: r1  r2  )rB  s   r6   r[   z lu_unpack_meta.<locals>.<lambda>F  s    kY[Yakkk r8   c                      dS )Nztorch.lu_unpack: LU_pivots is expected to be a contiguous tensor of torch.int32 dtype.
Note: this function is intended to be used with the output produced by torch.linalg.lu_factorr<   r<   r8   r6   r[   z lu_unpack_meta.<locals>.<lambda>K  s    p r8   r  r   r   )	rL   r]   r   rT   r=  r   r   r   r   )rB  r(  rO  rP  r@  r^  r  r]  r<  r  r  s   `          r6   lu_unpack_metarT  <  s?    
L
1kkkk    
LEK' 	
 	
 	
 NNEb	Ab	AAq		AE"I LLLL! b	LLb	b	LLLL!LL!a7Nr8   modec                 ~      dk    rd}d}n. dk    rd}d}n# dk    rd}d}nt          j        d fd           ||fS )NreducedTcompleteFrc                      d  dS )Nzqr received unrecognized mode 'z=' but expected one of 'reduced' (default), 'r', or 'complete'r<   )rU  s   r6   r[   z _parse_qr_mode.<locals>.<lambda>s  s"    O$ O O O r8   rL   r]   )rU  	compute_qrW  s   `  r6   _parse_qr_moder]  f  s~    y							   	
 	
 	
 gr8   QRrW  c                 T   t          | d           t          | d           t          |          \  }}| j        d         }| j        d         }t	          ||          }|rXt          | j                  }|r|n||d<   |                     |          }|                    |t          |d                     n|                     dg          }t          | j                  }	|s|s|n||	d<   |                     |	          }
|
                    |	t          |	d                     ||
fS )Nz	linalg.qrr  r   Fr  r   )	r  r   r]  r   r   r   r   r   r   )r  rU  r\  reduced_moder^  r  r]  Q_shaper^  R_shaper_  s              r6   linalg_qr_metard  {  s+    ![!!!1k***,T22I|	A	AAq		A qw--'.aaQKK  	g:7eTTTUUUUKK 17mmG#;9;!!!GBK	GAMM'6w%PPPQQQa4Kr8   sign	logabsdetc                    t          | d           t          | dd           | j        }|                     |d d                   }|                     |d d         t	          | j                            }t          j        |t          |d          | j        | j	                  }|                     |d d         t          j
                  }||||fS )Nzlinalg.slogdetFr  r   r   r   )r   r   r   r   rV   rT   rL   r!  r   rx   r=  )r  r   re  rf  rB  r(  s         r6   _linalg_slogdetrh    s     a)***1.666GE;;uSbSz""DE#2#Joag.F.FGGI		*5%88gx	
 
 
B [[ss5;[77FB&&r8   full_matrices
compute_uvdriverc                    t          | d           t          | d           t          | j        d d                   }| j        d         }| j        d         }t	          ||          }|r|||r|n|gz   }|                     |          }	|	                    |t          |d                     ||r|n||gz   }
|                     |
          }t          |           dk    }|                    |
t          |
|                     n,|                     dg          }	|                     dg          }|                     ||gz   t          | j
                            }|	||fS )	Nz
linalg.svdr  r   Fr  r   r   r   )r  r   r   r   r   r   r   r   r   rV   rT   )r  ri  rj  rk  r   r^  r  r]  U_shaper  V_shapeVis_cudaSs                 r6   _linalg_svd_metarr    sl    !\"""1l+++agcrcl##J	A	AAq		A #<111==KK  	g:7eTTTUUU] 91==KK  
 a..F*	g:7gVVVWWWW KKKK 	
J!$OAG,D,DEEAa7Nr8   arg1arg2c                 V   | j         d d         }|j         d d         }t          ||          }t          |          }||                     d          |                     d          gz  }t          |          }||                    d          |                    d          gz  }||fS )Nr  r   )r   r%   r   r   )rs  rt  arg1_batch_sizesarg2_batch_sizesexpand_batch_portionarg1_expand_sizearg2_expand_sizes          r6   r8  r8    s    
 z#2#z#2#,-=?OPP0112		"660112		"66---r8   c                     |rt          | ||           t          | |          \  }}|| j        k    r| n|                     |          }||j        k    r|n|                    |          }||fS r3   )r  r8  r   expand)rs  rt  r  ry  rz  arg1_broadcastedarg2_broadcasteds          r6   r  r    s      1tT4000)EdD)Q)Q&& !DJ..DKK@P4Q4Q  !DJ..DKK@P4Q4Q  ---r8   r   c                 v    | j         d d         }|j        dk    p| j        dz
  |j        k    o
|j         |k    }|S )Nr   r   )r   r   )r   r   expected_batched_rhs_shapevector_cases       r6   linalg_solve_is_vector_rhsr    sK    !&SbS!1*/ 
Q%*$R8R)R  r8   )r   r  r  rB  r(  r)  c                    t           d           t          j         j        j        k     fd           t	                     }|r                    d          n}	t           |	|d           t          |	           \  }
}t          j        |p| d            |r
|
d d         n|
}t          j        |t          ||           j        j
                  } j        }t          j        |t          |d           j         j
                  }                     |d d         t          j                  }                     |d d         t          j                  }||||f}||||f}t          d	 |D                       rjt          ||          D ]Y\  }}t!          ||j                   |                    |j        |                                           t'          ||d
           Z|S )Nzlinalg.solvec                  *    d j          dj          dS )NzKlinalg.solve: Expected A and B to have the same dtype, but found A of type rI  r1  r   )r  r  s   r6   r[   z"_linalg_solve_ex.<locals>.<lambda>  s0    9w9 9'(w9 9 9 r8   r   c                      dS )Nzlinalg.solve: Vector broadcasting of the left hand side is not supported for left=False. In this case linalg.solve is equivalent to B / A.squeeze(-1)r<   r<   r8   r6   r[   z"_linalg_solve_ex.<locals>.<lambda>  s    K r8   r   Fr   r  c              3      K   | ]}|d uV  	d S r3   r<   rB   rC   s     r6   ri   z#_linalg_solve_ex.<locals>.<genexpr>/  s&      
&
&Q1D=
&
&
&
&
&
&r8   )	copy_fromcopy_toexact_dtype)r   rL   r]   rT   r  	unsqueezer  r8  r!  r   rx   r   r   r=  allzipr!   r   r   r#   )r  r  r   r  r  rB  r(  r)  r  B_B_broad_shaperI   result_shaperesult_r   LU_pivots_info_r   resrY  os   ``                    r6   _linalg_solve_exr    s5    1n---	L	17	
 	
 	
 	
 	
   -Q22K'	.RQBaT>2223B::M1	LK	
 	
   *5G="%%-L!*<TBBgx	  G GE

*5%88gx	  C kk%*EKk88GKKcrc
%+K66E2vt
$CC%
(C

&
&#
&
&
&&& FSMM 	F 	FDAqa)))MM!'188::...QuEEEEEJr8   )r   unitriangularr   r  r   c                
   ||                      dg          }t          |t                    sJ t          | ||d           t	          || d           \  }}|                    dd                                          o|                                }|rt          ||j	                  }nYt          ||j	                  rD|                    |                    dd          j	                   |                    dd           |S )Nr   zlinalg.solve_triangularr  r   )r   rf   r   r  r  r  rZ  is_conjr!   r   r"   r   
transpose_)	r  r  r  r   r  r   r  A_avoid_copy_As	            r6   linalg_solve_triangular_metar  9  s     {kk1#c:&&&&&aD";<<<.q!T::FB<<B''5577HBJJLLL #RX..  RX.. 	#KKR,,2333NN2r"""Jr8   XM)r  r  c                     t          j         j        dk     fd           t          j        j        dk    fd           t           d           j        t           j        k    rvt                     \  }}t          j        |t          |d           j	         j
                  }t          j        |t          |d          j	        j
                  }nkj        t           j        k    sj        t           j        k    r+t          j                   }                     dg          }nt          j        dd	            ||fS )
Nr   c                      d j          dS )NzMtorch.triangular_solve: Expected b to have at least 2 dimensions, but it has r  r  r   s   r6   r[   z'triangular_solve_meta.<locals>.<lambda>^  s!    9)9 9 9 r8   c                      d j          dS )NzMtorch.triangular_solve: Expected A to have at least 2 dimensions, but it has r  r  r  s   r6   r[   z'triangular_solve_meta.<locals>.<lambda>e  r/  r8   triangular_solveFr  r   r   c                      dS )Nz+triangular_solve: Got an unexpected layout.r<   r<   r8   r6   r[   z'triangular_solve_meta.<locals>.<lambda>  s    $Q r8   )rL   r]   r   r  rw   stridedr8  r!  r   rT   rx   
sparse_csr
sparse_bsrr   r   )	r   r  r  r  r  self_broadcast_sizeA_broadcast_sizesolutioncloned_coefficients	   ``       r6   triangular_solve_metar  S  s    
L	Q	
 	
 	
 	
   
L	!	
 	
 	
 	
   4$6777x5=  0LTST0U0U--&$./BeTTT*;	
 
 
 #0!./?5QQQ'8	
 
 
 
U%	%	%U5E)E)E#D))!^^QC00UQQRRR'''r8   c                 ~   t          | d           t          | d           |                     | j        d d                   }|                     | j                  }|                    | j        t          | j        d                     |                     | j        d d         t          j                  }|||fS )Nz
linalg.detr  Fr  r   r   r%  )r  detrB  r(  s       r6   _linalg_det_metar    s    a&&&1l+++
++agcrcl
#
#C	
QW		BNN1775QQQRRR[["U[[99FF?r8   c                     t          j         j        dk    d            t          j        j        dk    d            |rdndt          j        j                 j        d         k    fd           t          j        j                  j        d         k    fd           t          j        j        d          j        d         k    d            t          j         j        j        z
  d	k     fd
           t          j         j        j        k     fd            j        dk    re j        d d         }j        d d         t          j        |k    fd           j        d d         t          j        |k    fd           t          j        j         j        k     fd           t          j        j         j        k     fd           t          d d           t          d d           t          j        j        t          j        d          j        j                  S )Nr   c                      dS )Nz3torch.ormqr: input must have at least 2 dimensions.r<   r<   r8   r6   r[   zormqr.<locals>.<lambda>      !V r8   c                      dS )Nz3torch.ormqr: other must have at least 2 dimensions.r<   r<   r8   r6   r[   zormqr.<locals>.<lambda>  r  r8   r  r   c                      d  dS )Ntorch.ormqr: other.shape[z0] must be greater than or equal to tau.shape[-1]r<   left_size_conditions   r6   r[   zormqr.<locals>.<lambda>  s    q,?qqq r8   c                      d  dS )Nr  z"] must be equal to input.shape[-2]r<   r  s   r6   r[   zormqr.<locals>.<lambda>  s    c,?ccc r8   c                      dS )NzHtorch.ormqr: tau.shape[-1] must be less than or equal to input.shape[-1]r<   r<   r8   r6   r[   zormqr.<locals>.<lambda>  r  r8   r   c                  (    dj          d j          S )Nz[torch.ormqr: Expected tau to have one dimension less than input, but got tau.ndim equal to r  r  r  s   r6   r[   zormqr.<locals>.<lambda>  r  r8   c                  (    dj          d j          S )Nzhtorch.ormqr: Expected other to have the same number of dimensions as input, but got other.ndim equal to r  r  r   r   s   r6   r[   zormqr.<locals>.<lambda>  s1    `+0:` `SXS]` ` r8   c                      d  S )NzWtorch.ormqr: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r<   r  s   r6   r[   zormqr.<locals>.<lambda>  r  r8   c                      d  S )NzYtorch.ormqr: Expected batch dimensions of other to be equal to input.shape[:-2], but got r<   )actual_batch_other_shapes   r6   r[   zormqr.<locals>.<lambda>  s    Q6NQ Q r8   c                  (    d j          dj          S )NzPtorch.ormqr: Expected input and tau to have the same dtype, but input has dtype z and tau has dtype r   r  s   r6   r[   zormqr.<locals>.<lambda>  s0    O#(;O OCF9O O r8   c                  (    d j          dj          S )NzRtorch.ormqr: Expected input and other to have the same dtype, but input has dtype z and other has dtype r   r  s   r6   r[   zormqr.<locals>.<lambda>  s0    S#(;S SEJ[S S r8   ztorch.ormqrr  r   Fr  r   )	rL   r]   r   r   rT   r  r!  r   rx   )	r   r  r   r   r  expected_batch_shaper  r  r  s	   ```   @@@r6   ormqrr    s    
L
aVV   
L
aVV   !%,"""	L'(CIbM9qqqq   
L'(EKO;cccc  
 
L	"R(ZZ  
 
L
SX"	
 	
 	
 	
 	
   
L
ej 	
 	
 	
 	
 	
   zA~~${3B3/!$3B3"&::   	
 	
 	
 $);ss#3 $(<<   	
 	
 	
 
L	U[ 	
 	
 	
 	
 	
   
Lu{"	
 	
 	
 	
 	
   M3u555M5%999[*5;%HHHk|	   r8   c                    t          j        t                    dz  k    fd            j        }|dz   k    }|}| }|r/t	          d|          D ]}|o                     |          dk    }n.t	          d|          D ]}|o                     |          dk    }t          j        |p| fd           d S )Nr   c                  4    dd z   dt                     S )Nzpadding size is expected to be r   z, but got: r   )r{   paddings   r6   r[   z,_padding_check_valid_input.<locals>.<lambda>  s"    T!c'TTc'llTT r8   r   r   c                  0    d dz    d dz    dj          S )N	Expected r   zD or r   zcD (batch mode) tensor with possibly 0 batch size and other non-zero dimensions for input, but got: r2  )r{   r   s   r6   r[   z,_padding_check_valid_input.<locals>.<lambda>  sC    Oa O OcAg O OAFO O r8   )rL   r]   r   r   r   r   )r   r  r{   	input_dimis_batch_modevalid_batch_modevalid_non_batch_moder   s   ```     r6   _padding_check_valid_inputr    s#   	LGCTTTTT  
 
I#'*M$,, Oq)$$ 	G 	GA/FEJJqMMQ4F	G q)$$ 	O 	OA#7#NEJJqMMQ<N   
L00	
 	
 	
 	
 	
    r8   c                   	
 d}dd} j         dk    r                     d          }dz  |dz  }t           |d           |\  	
                     |          }                               	z   
z   |r%t          j        	k     o
k      	
fd           t          j        dk    fd            j         dk    r                     |f          S                      ||f          S )Nr   r   r.   r   c                  *    d d d  dj          S NzcArgument #4: Padding size should be less than the corresponding input dimension, but got: padding (rm   ) at dimension 
 of input r2  dim_wr   pad_lpad_rs   r6   r[   z_pad1d_common.<locals>.<lambda>"  M    c%*c c.3c cDIc cUZU`c c r8   c                      d  d S )Nz
input (W: z%) is too small. Calculated output W: r<   )input_woutput_ws   r6   r[   z_pad1d_common.<locals>.<lambda>*  s    UWUU8UU r8   r   )r   r   r  rL   r]   r   )r   r  is_reflection	dim_planenbatchnplaner  r  r  r  r  s   `     @@@@@r6   _pad1d_commonr    sI   IEFzQA
Q	ug15555LE5ZZ	""FjjG&H 
GO/      	
 	
 	
 
LAUUUUU  
 zQ12229:::r8   c                 &    t          | |d          S NTr  )r  r   r  s     r6   meta_reflection_pad1dr  3       t<<<<r8   c                 ~     t          j         j        t           j        k     fd           t	           |d          S )Nc                  >    d j                                          dS )Nz)"replication_pad1d" not implemented for ''rT   __str__r   s   r6   r[   z(meta_replication_pad1d.<locals>.<lambda>>  !    Xek>Q>Q>S>SXXX r8   Fr  )rL   r]   rT   boolr  r  s   ` r6   meta_replication_pad1dr  9  H     
Luz!XXXX   u====r8   c                    d|s't          j        t          |          dk    d            j        dk    rdz  |\                                }|z   z   |r%t          j        |k     o|k     fd           t          j                                       k     fd                               j                  S )Nr   r   c                      dS )Nz padding size is expected to be 2r<   r<   r8   r6   r[   z(_pad1d_backward_common.<locals>.<lambda>F  s    0R r8   r.   c                  *    d d d  dj          S r  r2  r  s   r6   r[   z(_pad1d_backward_common.<locals>.<lambda>S  r  r8   c                  :    d d                                 S Nz(grad_output width unexpected. Expected: , Got: r   r  grad_outputr  s   r6   r[   z(_pad1d_backward_common.<locals>.<lambda>[  '    e8eeKL\L\]bLcLcee r8   rL   r]   r   r   r   r   r   )	r  r   r  r  r  r  r  r  r  s	   ``   @@@@r6   _pad1d_backward_commonr  C  s   E TS\\Q&(R(RSSSzQ
LE5jjG&H 
GO/      	
 	
 	
 
LK$$U+++eeeeee  
 ??5;'''r8   
grad_inputc                 (    t          | ||d          S r  r  r  r   r  s      r6   meta_reflection_pad1d_backwardr  a  s     "+ugTRRRRr8   c                 (    t          | ||d          S )NFr  r  r   s      r6   meta_replication_pad1d_backwardr  g  s     "+ugUSSSSr8   c                   	
 ddd}d}t           |d            j        }|dk    r$                     d          }dz  dz  |dz  }|\                       |          }                               	                               
	z   z   
z   z   |rJt          j        
k     o
k      fd           t          j        	k     o	k      fd           t          j        dk    pdk    	
fd            j        d	k    r                     |f          S                      ||f          S )
Nr   r   r   r      c                  *    d d d  dj          S r  r2  r  s   r6   r[   z_pad2d_common.<locals>.<lambda>  r  r8   c                  *    d d d  dj          S NzcArgument #6: Padding size should be less than the corresponding input dimension, but got: padding (rm   r  r  r2  dim_hr   pad_bpad_ts   r6   r[   z_pad2d_common.<locals>.<lambda>  r  r8   c                       d  d d d S )Nz
input (H:  W: z%) is too small. Calculated output H: r<   )input_hr  output_hr  s   r6   r[   z_pad2d_common.<locals>.<lambda>  s?    = = =g = =$,= =2:= = r8   r.   r  r   r   rL   r]   r   )r   r  r  
dim_slicesr  r   r  r
  r  r  r  r  r  r  r  r  r  s   `      @@@@@@@@@@r6   _pad2d_commonr  m  s   EEJFug15555:DqyyA

a
!(E5%ZZ
##FjjGjjG&H&H 
GO/      	
 	
 	
 	GO/      	
 	
 	
 
LA&Q	
 	
 	
 	
 	
 	
 	
   zQ(;<<<(CDDDr8   c                 &    t          | |d          S r  )r  r  s     r6   meta_reflection_pad2dr    r  r8   c                 ~     t          j         j        t           j        k     fd           t	           |d          S )Nc                  >    d j                                          dS )Nz)"replication_pad2d" not implemented for 'r  r  r  s   r6   r[   z(meta_replication_pad2d.<locals>.<lambda>  r  r8   Fr  )rL   r]   rT   r  r  r  s   ` r6   meta_replication_pad2dr    r  r8   c                     ddd}|j         }|                                dk    rdz  dz  |dz  }|\  }}}}|         }	|         }
|	|z   |z   |
|z   |z   t          j                                       k     fd           t          j                                       k     fd           |                    |j                   S )Nr   r   r   r  c                  :    d d                                 S r  r   r  s   r6   r[   z%meta_pad2d_backward.<locals>.<lambda>  r  r8   c                  :    d d                                 S Nz)grad_output height unexpected. Expected: r  r   r
  r  r  s   r6   r[   z%meta_pad2d_backward.<locals>.<lambda>  '    fHff[M]M]^cMdMdff r8   )r   r{   rL   r]   r   r   )r  r   r  r  rZ   r  r  r  r  r  r  r
  r  r  r  s   `          @@@@r6   meta_pad2d_backwardr    s    EEIJxxzzQ

Q	!(E5%GG&H&H	LK$$U+++eeeeee   
LK$$U+++ffffff   >>$*%%%r8   c          	      R   	
 d	ddd}t           |d            j        dk    }|r)                     d          }	dz  	dz  dz  |dz  }|\                       |          }                               
                                                    	          
z   z   z   z   z   z   |rot          j        k     ok     	 fd           t          j        k     ok      fd           t          j        
k     o
k      fd	           t          j        dk    pdk    pdk    
fd
           |r                     ||f          S                      |f          S )Nr.   r   r   r   r      c                  *    d d d  dj          S r  r2  r  s   r6   r[   z_pad3d_common.<locals>.<lambda>  r  r8   c                  *    d d d  dj          S r  r2  r	  s   r6   r[   z_pad3d_common.<locals>.<lambda>  r  r8   c                  *    d d d  dj          S )NzcArgument #8: Padding size should be less than the corresponding input dimension, but got: padding (rm   r  r  r2  )dim_dr   pad_bkpad_fs   r6   r[   z_pad3d_common.<locals>.<lambda>  sM    d%*d d.4d dEJd dV[Vad d r8   c                  ,    d  d d d d d S )Nz
input (D:  H: r  z%) is too small. Calculated output D: r<   )input_dr  r  output_dr  r  s   r6   r[   z_pad3d_common.<locals>.<lambda>  se    K K Kg K K7 K K$,K K2:K K@HK K r8   r  )r   r  r  r  
batch_moder  r  r%  r
  r  r*  r  r  r+  r  r  r  r&  r'  r  r  r  s   `      @@@@@@@@@@@@@@@r6   _pad3d_commonr-    s   EEEIug15555qJ A


Q	07-E5%vZZ	""FjjGjjGjjG'H&H&H 
GO/      	
 	
 	
 	GO/      	
 	
 	
 	GO0 0      	
 	
 	
 
LA7Q7(a-	
 	
 	
 	
 	
 	
 	
 	
 	
    G(HMNNN(HEFFFr8   c                 &    t          | |d          S r  )r-  r  s     r6   meta_reflection_pad3dr/    r  r8   c                 ~     t          j         j        t           j        k     fd           t	           |d          S )Nc                  >    d j                                          dS )Nz)"replication_pad3d" not implemented for 'r  r  r  s   r6   r[   z(meta_replication_pad3d.<locals>.<lambda>"  r  r8   Fr  )rL   r]   rT   r  r-  r  s   ` r6   meta_replication_pad3dr2    r  r8   c                     t          j        t          |          dk    d            |j        dk    sJ  j        |j        k    sJ ddd|j        dk    rdz  dz  dz  |\  }}}}}}|                              }	|                              }
|                              }|	|z   |z   |
|z   |z   ||z   |z   t          j                                       k     fd           t          j                                       k     fd           t          j                                       k     fd	           |                    |j                  S )
N   c                      dS )Nz padding size is expected to be 6r<   r<   r8   r6   r[   z%meta_pad3d_backward.<locals>.<lambda>1  s    ,N r8   r.   r   r   r!  c                  :    d d                                 S r  r   r  s   r6   r[   z%meta_pad3d_backward.<locals>.<lambda>I  r  r8   c                  :    d d                                 S r  r   r  s   r6   r[   z%meta_pad3d_backward.<locals>.<lambda>M  r  r8   c                  :    d d                                 S )Nz(grad_output depth unexpected. Expected: r  r   )r%  r  r+  s   r6   r[   z%meta_pad3d_backward.<locals>.<lambda>Q  r  r8   r  )r  r   r  r  r  r  r  r'  r&  r*  r  r  r%  r
  r  r+  r  r  s   `           @@@@@@r6   meta_pad3d_backwardr9  '  s    
LW"$N$NOOO:>>>>uz))))EEEzQ


07-E5%vjjGjjGjjG'H&H&H	LK$$U+++eeeeee   
LK$$U+++ffffff   
LK$$U+++eeeeee  
 ??5;'''r8   r   pc                 j   t          j        |                                 d            |                     d          }|dk    r4|                     dg                              t           j                  S |                     ||dz
  z  dz  f                              t           j                  S )Nc                      dS )Nz(_pdist_forward requires contiguous inputr<   r<   r8   r6   r[   z%meta__pdist_forward.<locals>.<lambda>[  s    &P r8   r   r   r   r   )rL   r]   rZ  r   r   r"  r  )r   r:  r  s      r6   meta__pdist_forwardr=  W  s     
LPP   			!AAvv~~qc""%%E4R%SSS~~qAE{a/122558 6 
 
 	
r8   gradpdistc                     t          j        |                                d            t          j        |                                d            t          j        |t           j                  S )Nc                      dS )Nz._pdist_backward requires self to be contiguousr<   r<   r8   r6   r[   z&meta__pdist_backward.<locals>.<lambda>j  s    &V r8   c                      dS )Nz/_pdist_backward requires pdist to be contiguousr<   r<   r8   r6   r[   z&meta__pdist_backward.<locals>.<lambda>m  s    'X r8   r   )rL   r]   rZ  r   r  )r>  r   r:  r?  s       r6   meta__pdist_backwardrC  f  ss     
LVV   
LXX   D0NOOOOr8   )rI  rH  c          
      J    ddl m}m}                     d          }                    d          }                    d          }	 |t	          j         | j        |||	f                              r                     |||	f           t	          j        	                                dk    d            t	          j        	                                dk    d            t          j        s7t	          j         j        j        cxk    o
j        k    nc  fd           j        }
j        |
d         |
d         t	          j        d         k    od         k    fd	                                                                           S )
Nr   )guard_or_truesym_eqr   r   r.   c                      dS Nzbatch1 must be a 3D tensorr<   r<   r8   r6   r[   zmeta_baddbmm.<locals>.<lambda>|      ,H r8   c                      dS Nzbatch2 must be a 3D tensorr<   r<   r8   r6   r[   zmeta_baddbmm.<locals>.<lambda>}  rI  r8   c                  8    dj          d j          dj          S )Nz+Input dtypes must be the same, got: input: z
, batch1: z
, batch2: r   )batch1batch2r   s   r6   r[   zmeta_baddbmm.<locals>.<lambda>  s*    ~$*~~X^Xd~~pvp|~~ r8   c            	      :    d d d d          d d          d	S Nz@Expected size for first two dimensions of batch2 tensor to be: [rm   z] but got: [r   r   ].r<   batch2_sizesbscontraction_sizes   r6   r[   zmeta_baddbmm.<locals>.<lambda>  sY    [[ [&[ [4@O[ [GSTU[ [ [ r8   )r  rE  rF  r   rL   sym_notr   r|  r]   r{   
exp_config&skip_dtype_check_in_meta_registrationsrT   r   )r   rM  rN  rI  rH  rE  rF  dim1dim2dim3batch1_sizesrS  rT  rU  s   ```        @@@r6   meta_baddbmmr]  r  s    LKKKKKKK;;q>>D;;q>>D;;q>>D}U]66$*tT46H#I#IJJKK /{{D$-..	L"$H$HIII	L"$H$HIII< 
J&,6666&,6666~~~~~~	
 	
 	
 <L<L	aB#A	LQ2E,q/5E"E	
 	
 	
 	
 	
 	
   >>$))++&&&r8   c                B    t          j        | t           j                  S r   re  r   r   s     r6   meta_bernoullir`    s     D0GHHHHr8         ?c                     | S r3   r<   r   r:  r   s      r6   meta_bernoulli_rd        Kr8   c                 B    t          j        | t           j                  S r   re  rc  s      r6   meta_bernoulli_prg    s     D0GHHHHr8   c                 *    t          j        |           S r3   rL   r   r_  s     r6   meta_poissonrj         D!!!r8   c                     t          j        |
|                                 k     d            t          j        | t           j                  }t          j        |           |fS )Nc                      dS )NzJError in fused_moving_avg_obs_fake_quant_cpu: ch_axis must be < self.dim()r<   r<   r8   r6   r[   z6meta__fused_moving_avg_obs_fq_helper.<locals>.<lambda>      \ r8   r   )rL   r]   r{   r   r  )r   observer_onfake_quant_onrunning_minrunning_maxscale
zero_pointaveraging_const	quant_min	quant_maxch_axisper_row_fake_quantsymmetric_quantmasks                 r6   $meta__fused_moving_avg_obs_fq_helperr|    s]      
L$((**\\   D
333DT""D))r8   c                 L   t          j        |                                 dk    d            t          j        |                                dk    d            | j        \  |j        \  t          j        k    fd           |                               S )Nr   c                      dS )Nza must be 2Dr<   r<   r8   r6   r[   zmeta_mm.<locals>.<lambda>      ~ r8   c                      dS )Nzb must be 2Dr<   r<   r8   r6   r[   zmeta_mm.<locals>.<lambda>  r  r8   c            	      "    d d  d d d	S )Nz/a and b must have same reduction dim, but got [rm   z] X [rQ  r<   )M1M2Nr<  s   r6   r[   zmeta_mm.<locals>.<lambda>  s-    [![[r[[PR[[VW[[[ r8   )rL   r]   r{   r   r   )abr  r  r  r<  s     @@@@r6   meta_mmr    s     
LA55666	LA55666GEArGEB	L
b[[[[[[[   ;;q!r8   c                      |r.t           fdt           j                  D                       S t          j         j                  S )Nc              3   >   K   | ]}|vrj         |         nd V  dS )r   Nr2  )rB   r   dimsr   s     r6   ri   z+_compute_reduction_shape.<locals>.<genexpr>  s5      UUqatmmTZ]]UUUUUUr8   )r\   r   r   rF   compute_reduction_output_shaper   )r   r  r  s   `` r6   r  r    sR     VUUUUUE$)DTDTUUUUUU/
DAAAr8   strc                     t          | t          j        j                  r| j        j        S t          | d          r1t          | j        d          r| j        j        dk    r| j        j        S dS )Nrx   ro   ru   r   )rf   rL   _subclasses
FakeTensorfake_devicero   hasattrrx   )r,  s    r6   r   r     sn    &%+677 	!&&!!FM6** M&((}!!vr8   input_tensorr   r  dilationis_transposedgroupsoutput_paddingc                 6   dt           dt           dt           dt           dt           dt           fd}dt           dt           dt           dt           dt           dt           dt           fd	}	|j        d
d          }
| j        d
d          |r||j        d         z  }n;|j        d         }|j        d         |z  | j        d         k    rt          d          | j        d         |gt          |t                    r|gt                    z  }n,t          |          dk    r|d         gt                    z  }t          |t                    r|gt                    z  }n,t          |          dk    r|d         gt                    z  }t          |t                    r|gt                    z  }n,t          |          dk    r|d         gt                    z  }d }|rXt          |t                    r|gt                    z  }n/t          |          dk    r|d         gt                    z  }n|}t          t                              D ]}|rH                     |	|         ||         ||         |
|         ||         ||                              L                     ||         ||         ||         |
|         ||                              ddlm	} t          j         |d d
d          D              fd           S )Nlnr:  r   r]  r  r/   c                 6    | d|z  z   ||dz
  z  z
  dz
  |z  dz   S )a  
        Formula to apply to calculate the length of some dimension of the output

        See: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html

        Args:
            ln: length of the dimension
            p: padding in that dim
            d: dilation in that dim
            k: kernel size in that dim
            s: stride in that dim
        Returns:
            The output length
        r   r   r<   )r  r:  r   r]  r  s        r6   _formulaz+calc_conv_nd_return_shape.<locals>._formula  s.     QU
Q!a%[(1,2Q66r8   r4   c                 <    | dz
  |z  d|z  z
  ||dz
  z  z   |z   dz   S )a  
        Formula to apply to calculate the length of some dimension of the output
        if transposed convolution is used.
        See: https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html

        Args:
            ln: length of the dimension
            p: padding in that dim
            d: dilation in that dim
            k: kernel size in that dim
            s: stride in that dim
            op: output padding in that dim

        Returns:
            The output length
        r   r   r<   )r  r:  r   r]  r  r4   s         r6   _formula_transposedz6calc_conv_nd_return_shape.<locals>._formula_transposed	  s2    " Q!|a!e#a1q5k1B6::r8   r   r   r   zInvalid channel dimensions)sym_orc                     g | ]}|d k    	S r   r<   r  s     r6   rE   z-calc_conv_nd_return_shape.<locals>.<listcomp>K	  s    ...1Q...r8   c                  @    dt                      ddd           dS )NzGiven input size per channel: z&. Calculated output size per channel: r   z. Output size is too small)r   )r  	ret_shapes   r6   r[   z+calc_conv_nd_return_shape.<locals>.<lambda>L	  s6     $d $ $/8}$ $ $ r8   )r   r   r!  rf   r   r   r   r   r  r  rL   r]   )r  r5  r   r  r  r  r  r  r  r  kernel_sizeout_channelsoutput_padding_listr   r  r  r  s                  @@r6   calc_conv_nd_return_shaper    s   7S 7S 7S 7S 7S 7S 7 7 7 7"; ; ; ; ; ; ;QT ; ; ; ;& ,qrr"Kabb!D =Q/|A<?V#|'9!'<<<;<<<#A&5I&'"" )CII%	V		)s4yy('7## +)c$ii'	W		1:,T*(G$$ -:D		)	X!		QK=3t99,/3 1ng.. 	1#1"2SYY">  A%%#1!#4"5D		"A"03t99   	##GAJQKN1I'* 	 	 	 	 a'!*hqk;q>6RS9UU    =<<<<<	L..	!"".../	$ 	$ 	$ 	$ 	$   r8   c                 \    t           j                            |           t           j        k    S r3   rL   _prims_commonr   channels_lasttens    r6   is_channels_lastr  T	  s!    44S99U=PPPr8   running_meanrunning_vartrainingexponential_average_factorepsilonc                 v     j         }||j         n|j         }	||j         n|j         }
 fd}                     |                               |                      }|r+                     |	          }                     |
          }n*                     d          }                     d          }|||fS )Nc                      t                     rt          j        S                      t          j                  rt          j        S t          j        S r   )r  rL   r  rZ  r   )r  s   r6   pick_memory_formatz2meta_miopen_batch_norm.<locals>.pick_memory_formatk	  sJ    L)) 	'&&%%E4K%LL 	+**&&r8   r   r   )r   r   r"  )r  r5  r7  r  r  r  r  r  r   save_mean_shapesave_var_shaper  r   	save_meansave_vars   `              r6   meta_miopen_batch_normr  X	  s     "I -9,Dl((&,O*5*A[&&v|N' ' ' ' ' 
 
 
+
+
.
.=O=O=Q=Q
.
R
RC 0 **?;;	)).99 **400	))$//	8##r8   c	           
            fd}	t           ||||||r|nd           }
d}d}                     |          dk    rd|
|<                        |
          }|                     |	                      }|S )Nc                  j   t                     dk    r+t                     st                    rt          j        S nt                     rt          j        S                      t          j                  rt          j        S                      t          j                  rt          j        S d S Nr   r   )r   r  rL   r  rZ  r   preserve_format)r  r5  s   r6   r  z%meta_conv.<locals>.pick_memory_format	  s    |$$..-- +1A&1I1I +**+  -- +**%%E4K%LL 	)**''e6K'LL 	)((	) 	)r8   r   r   r   )r  r   r   r"  )r  r5  r7  r   r  r  r  r  r  r  	shape_outinput_channels_dimoutput_channels_dimr   s   ``            r6   	meta_convr  ~	  s    
) 
) 
) 
) 
) 
) *'1T	 	I +,,11)*	%&

 
 
+
+C
&&1133&
4
4CJr8   mkldnnc
           
          t          | ||||d|g           }
|                     |
          }t          j        }|                                 dk    rt          j        }|                    |          }|S )NFr!  r   )r  r   rL   r  r{   channels_last_3dr"  )r  r5  r7  r  r   r  r  attrscalars	algorithmr  r   out_memory_formats                r6   meta_mkldnn_convolution_defaultr  	  s{     .&&'8UFB
 
	 $$Y//!/"" % 6ff#4f55
r8   c                 f    |                      g | j        d d         |j        d         R           S Nr   r   r   r   )r  r5  r7  r  r  r  s         r6   meta_linear_pointwise_defaultr  	  s8     %%&Q(:3B3(?&Qa&Q&QRRRr8   mklc                 f    |                      g | j        d d         |j        d         R           S r  r  )r  packed_weightorig_weightr7  r   s        r6   meta_mkl_linearr  	  s>    ))@,$SbS)@;+<Q+?@@  r8   onednnc           
         t          | ||||	d|
d           }|| j        }|t          j        t          j        t          j        t          j        t          j        fv sJ |                     ||          }t          |          dv s
J d            t          j
        t          j        t          j        dt          |                   }|                    |          }|S )NFr   )r.   r  r!  z-Expect output to be 3d/4d/5d for conv1d/2d/3dr   )r  rT   rL   rV  rX  uint8r<  rY  r   r   r   r  r  r"  )rC   x_scalex_zpww_scalew_zpr7  r   r  r  r  output_scaleoutput_zero_pointoutput_dtyper  r  r  r  r   formats                       r6   meta_qconv_pointwiser  	  s    * .	
 	
	 7LMNKJ 
 
 
 
 
 kk)<k889~~***; +** &"%
 
 i..	
 ff6f**
r8   c                     |dk    sJ |S )Nsumr<   )rC   r  r  r  r  r  accumr7  r   r  r  r  r  r  r  accum_scaleaccum_zero_pointbinary_op_namerH  unary_op_nameunary_op_argsunary_op_algorithms                         r6   meta_qconv2d_pointwise_binaryr  
  s    2 &&&&r8   c                     t          | j                  }|j        d         |d<   |	t          j        t          j        t          j        t          j        t          j        fv sJ |                     ||	          }|S )Nr   r   r   )	r   r   rL   rV  rX  r<  r  rY  r   )rC   r  r  r  r  r  r7  r  r  r  post_op_namepost_op_argspost_op_algorithmr_  r   s                  r6   meta_qlinear_pointwiser  ,
  ss    " AG}}71:RMNJK 
 
 
 
 
 kk,lk;;
r8   c                    |dk    r|S t          | j                  }|j        d         |d<   |
t          j        t          j        t          j        t          j        t          j        fv sJ |                     ||
          }|S )Nr  r   r   r   )	r   r   rL   rV  rX  r  r<  rY  r   )rC   r  r  r  r  r  x_2r7  r  r  r  x2_scalex2_zpr  rH  r  r  r  r_  r   s                       r6   meta_qlinear_pointwise_binaryr  J
  s    , U""JAG}}71:RMNKJ 
 
 
 
 
 kk,lk;;
r8   c                 x    t          | j                  }|j        d         |d<   |                     |          }|S )Nr   r   )r   r   r   )rC   r  r7  r_  r   s        r6   meta_linear_dynamic_fp16r  o
  s7     AG}}71:Rkk,''
r8   	quantizedr<   r   r   c                 4   t          | |||||          \  }}}|                                 dk    r|                     d          nd}	t          j        }
|                                 dk    r|||g}n|	|||g}t          j        || j        | j        |
          S Nr  r   r.   rv  )#max_pool2d_checks_and_compute_shaper{   r   rL   r  r   rT   rx   r   r  r   r  r  	ceil_modenInputPlaneoutputHeightoutputWidthr  r   r   s               r6   meta_quantized_max_pool2dr   
  s     0;9
 
		
 $)99;;!#3#3B+99;;!{;DDK{CD{+<'	
 
 
 	
r8   c                 ,   t          j        |                                 dk    d|                                  d           t          j        |                                dk    d|                                 d           t          j        | j        t           j        t           j        t           j        fv d| j                    t          j        |j        t           j        k    d|j                    t          j        |j        t           j        k    d|j                    t          j        |j        | j        k    d|j                    | 	                    | 
                    d	          |
                    d	          | j        
          S )Nr   zx must be a 2D tensor, got Dzw must be a 2D tensor, got #expected x to be f32/f16/bf16, got expected w to be uint8, got z q_group_size must be int64, got z5q_scale_and_zeros must have the same dtype as x, got r   r   )rL   r]   r{   rT   rV  rW  rX  r  r   r   r   rC   r  q_group_sizeq_scale_and_zeross       r6   meta_int4mm_packed_weight_cpur  
  sW   QUUWW\#K#K#K#KLLLQUUWW\#K#K#K#KLLLGu}enEE;!';;	
 	
 	
 	QW+-UAG-U-UVVV%+-C|/ACC	
 	
 	
 	#qw.]DUD[]]	
 	
 	
 {{166!99affQiiqw{???r8   c                      t          j                                         k    o j                 k     fd           d S )Nc                  n    d  d d dd                                  d dj                  z   S )NzExpected a tensor of dimension z and tensor.size[z] == rm   zbut got : dimension z] = r{   r   )r{   dim_sizer   r,  s   r6   r[   z check_dim_size.<locals>.<lambda>
  sU    _#____W[___
f
f
f
f
ffl[cNd
f
fg r8   )rL   r]   r{   r   )r,  r{   r  r   s   ````r6   check_dim_sizer  
  so    	L

>X 6$ >	g 	g 	g 	g 	g 	g 	g    r8   c                     d } |d|          \  }}	t          j        t          |          dv d            t          j         j        t           j        t           j        t           j        t           j        fv fd           t          |          dk    r||	}}
n3t          |          dk    r|d         |d         }}
n |d|          \  }
} |d	|          \  }}t          j        |d u p|dk    d
                                             dk    r 	                    d          nd} 	                    d          } 	                    d          } 	                    d          }t          ||||
d|          }t          ||	||d|          }t          j                   }t           ||	|
|||dd||||||                                            dk    r|||g}n||||g}t          j        | j         j        |          S )Nc                      t          j        t          |          dv  fd           |d         }t          |          dk    r|n|d         }||fS )Nr   r   c                      d  dS )Nzavg_pool2d: 4 must either be a single int, or a tuple of two intsr<   r  s   r6   r[   z1meta_avg_pool2d.<locals>.unpack.<locals>.<lambda>
      ]4]]] r8   r   r   rL   r]   r   r  r  HWs   `   r6   unpackzmeta_avg_pool2d.<locals>.unpack
  `    HH]]]]	
 	
 	
 FSQAACF!tr8   r  r   r   r   c                      dS NzOavg_pool2d: stride must either be omitted, a single int, or a tuple of two intsr<   r<   r8   r6   r[   z!meta_avg_pool2d.<locals>.<lambda>
      a r8   c                  >    d j                                          dS )Nz""avg_pool2d" not implemented for 'r  r  r  s   r6   r[   z!meta_avg_pool2d.<locals>.<lambda>
  !    Qu{7J7J7L7LQQQ r8   r   r   r   r  c                      dS Nzdivisor must be not zeror<   r<   r8   r6   r[   z!meta_avg_pool2d.<locals>.<lambda>
      * r8   r  r  r  r   r.   rv  )rL   r]   r   rT   r  uint16uint32uint64r{   r   pooling_output_shaperF   r   pool2d_shape_checkr   rx   )r   r  r   r  r  count_include_paddivisor_overrider  kHkWdHdWpadHpadWr  r  inputHeight
inputWidthr  r  r   r   s   `                     r6   meta_avg_pool2dr4  
  sR      VM;//FB	LFy aa   
LEKu|U\RRQQQQ   6{{aRB	V		F1IB&))B	7++JD$	LD 9$4$9**  
  %yy{{a//UZZ^^^QF**R..K**R..KBJ'Rr1iPPL&z2tRINNK/66M



		  $ yy{{a\;7\;?;k|#	   r8   c                     t          | ||||||dd|	|
||||           |                                 }|	}t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr   r.   r   )r)  r{   r  )r   
gradOutputr  r,  r-  r.  r/  r0  r1  r  r2  r3  r  r  
mem_formatr   nOutputPlanes                    r6   avg_pool2d_backward_shape_checkr9    s    " 



		  $ 99;;DL:tTAX|<<<:tTAX|<<<:tTAX{;;;;;r8   c                 j   t          j        t          |          dk    pt          |          dk    d            |d         }t          |          dk    r|n|d         }	t          j        t          |          dk    p%t          |          dk    pt          |          dk    d            t          |          dk    r|n|d         }
t          |          dk    r|	nt          |          dk    r|
n|d         }t          j        t          |          dk    pt          |          dk    d            |d         }t          |          dk    r|n|d         }t          j        |d u p|dk    d            |j        }|                                dk    r|d	         nd}|d
         }|d         }|d         }t          ||||
d|          }t          ||	||d|          }t          j        |          }t          || |||	|
|||||||||           t          j	        ||j
        |j        |          S )Nr   r   c                      dS )NzKavg_pool2d: kernel_size must either be a single int, or a tuple of two intsr<   r<   r8   r6   r[   z*meta_avg_pool2d_backward.<locals>.<lambda>F  s    ] r8   r   c                      dS r  r<   r<   r8   r6   r[   z*meta_avg_pool2d_backward.<locals>.<lambda>L  r  r8   c                      dS )NzGavg_pool2d: padding must either be a single int, or a tuple of two intsr<   r<   r8   r6   r[   z*meta_avg_pool2d_backward.<locals>.<lambda>R  s    Y r8   c                      dS r"  r<   r<   r8   r6   r[   z*meta_avg_pool2d_backward.<locals>.<lambda>Y  r#  r8   r  r  r$  r  r   rv  )rL   r]   r   r   r{   r(  rF   r   r9  r   rT   rx   )gradOutput_r   r  r   r  r  r*  r+  r,  r-  r.  r/  r0  r1  
input_sizer  r  r2  r3  r  r  r7  s                         r6   meta_avg_pool2d_backwardrA  8  sz    
LKA6[!1!1Q!6]]   
QB;1$$+a.B	LFq@CKK1,@Fq0@aa   6{{aVAYB6{{a3v;;!+;+;RRB	LG.S\\Q.YY   1:Dw<<1$$44'!*D	LD 9$4$9**  
 J$yy{{a//Z^^QFR.KR.KBJ'Rr1iPPL&z2tRINNK,U33J#



  $ ;k| 	   r8   c                     t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }	t          j        | pt          |          dv d            t          j         j        t           j        t           j        t           j        t           j        fv fd           |s|n|d         }
|s|nt          |          dk    r|
n|d         }|s|	nt          |          dk    r|
n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j         j        d	v d
            t          j        | p|dk    d             	                    d          } 	                    d          } 	                    d          } 	                    d          } 	                    d          }t          ||||
d|          }t          ||||d|          }t          ||	||d|          }t           ||||	|
|||||ddd||||||dd            j        dk    r                     ||||f          S                      |||||f          S )Nr   r.   c                      dS NzFavg_pool3d: kernel_size must be a single int, or a tuple of three intsr<   r<   r8   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>      X r8   r   r   r   c                      dS NzJavg_pool3d: stride must be omitted, a single int, or a tuple of three intsr<   r<   r8   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>  rn  r8   c                  >    d j                                          dS )Nz""avg_pool3d" not implemented for 'r  r  r  s   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>  r   r8   c                      dS NzBavg_pool3d: padding must be a single int, or a tuple of three intsr<   r<   r8   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>      T r8   r  r!  c                      dS Nz9non-empty 4D or 5D (batch mode) tensor expected for inputr<   r<   r8   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>      K r8   c                      dS r"  r<   r<   r8   r6   r[   z!meta_avg_pool3d.<locals>.<lambda>  r#  r8   r  r$  r  r   zavg_pool3d()T)check_input_sizer  )rL   r]   r   rT   r  r%  r&  r'  r   r   r(  pool3d_shape_checkr   )r   r  r   r  r  r*  r+  kTr,  r-  dTr.  r/  padTr0  r1  r  nslicesitimeiheightiwidthotimeoheightowidths   `                       r6   meta_avg_pool3dr^    s2    
LKF"XX   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+\\   
LEKu|U\RRQQQQ   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LGTT   1:Dw<<1$$44'!*Dw<<1$$44'!*D	L
fKK  
 
L5 0A 5**  
 ZZ]]FjjnnGJJrNNEjjnnGZZ^^F D"aCCE"7Bb!YGGG!&"dB9EEF





			-   2 zQ@AAAHIIIr8   c                    t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }	t          |          dk    r|n|d         }
t          j        | pt          |          dv d            |s|n|d         }|s|	nt          |          dk    r|n|d         }|s|
nt          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        |j        dv d	            t          j        | p|dk    d
            |                    d          }|                    d          }|                    d          }|                    d          }t          ||||d|          }t          ||	||d|          }t          ||
||d|          }t          || |||	|
||||||||||||d           |                    |j                  S )NrC  c                      dS rE  r<   r<   r8   r6   r[   z*meta_avg_pool3d_backward.<locals>.<lambda>  rF  r8   r   r   r   c                      dS rH  r<   r<   r8   r6   r[   z*meta_avg_pool3d_backward.<locals>.<lambda>  rn  r8   c                      dS rK  r<   r<   r8   r6   r[   z*meta_avg_pool3d_backward.<locals>.<lambda>  rL  r8   rM  c                      dS rO  r<   r<   r8   r6   r[   z*meta_avg_pool3d_backward.<locals>.<lambda>  rP  r8   c                      dS r"  r<   r<   r8   r6   r[   z*meta_avg_pool3d_backward.<locals>.<lambda>  r#  r8   r  r$  r  r   zavg_pool3d_backward())	rL   r]   r   r   r   r(  avg_pool3d_backward_shape_checkr   r   )r  r   r  r   r  r  r*  r+  rT  r,  r-  rU  r.  r/  rV  r0  r1  rW  rX  rY  rZ  otime_for_shape_checkoheight_for_shape_checkowidth_for_shape_checks                           r6   meta_avg_pool3d_backwardri    s    
LKF"XX   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+\\   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LGTT   1:Dw<<1$$44'!*Dw<<1$$44'!*D	L
fKK  
 
L5 0A 5**  
 jjnnGJJrNNEjjnnGZZ^^F0D"aSS27Bb!YWW1&"dB9UU#





'  , ??5;'''r8   c                     t          j         j        dk    p
 j        dk     fd            j        d d         t	          |          z   }t          j                   }t          j        | j         j	        |          S )Nr.   r  c                      d j          S )Nz"Expected 3D or 4D tensor, but got r2  r   s   r6   r[   z*meta_adaptive_avg_pool2d.<locals>.<lambda>/      ATZAA r8   r  rv  )
rL   r]   r   r   r\   rF   r   r   rT   rx   )r   output_sizer_  r   s   `   r6   meta_adaptive_avg_pool2drn  +  s    	L	Q($)q.AAAA   :crc?U;%7%77L/55M ;j{#	   r8   c                      t          j         j        dk    p
 j        dk     fd                                 j        d d         t          |          z             S )Nr  r!  c                      d j          S )Nz"Expected 4D or 5D tensor, but got r2  r   s   r6   r[   z*meta_adaptive_avg_pool3d.<locals>.<lambda>A  rl  r8   r$  )rL   r]   r   r   r   r\   )r   rm  s   ` r6   meta_adaptive_avg_pool3drq  =  sb    	L	Q($)q.AAAA   >>$*SbS/E+,>,>>???r8   c                      j         }t          d|          D ]2t          j                                       dk     fd           3t          j        |dk    p|dk    fd           t          j        j         j        k     fd           t          j        }t                    rt          j        }	                    j
                                      |          S )	Nr   r   c                       d j          d dS )Nz{adaptive_avg_pool2d_backward(): Expected grad_output to have non-zero                       size for non-batch dimensions,  with dimension  being emptyr2  )grad_outr   s   r6   r[   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>L  s1     f6>nf fVWf f f r8   r.   r  c                      d j          S )NzBadaptive_avg_pool2d_backward(): Expected 3D or 4D tensor, but got r2  r   s   r6   r[   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>Q  s    aUYU_aa r8   c                  (    dj          d j          S Nexpected dtype z! for `grad_output` but got dtype r   )rv  r   s   r6   r[   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>U  s    _$*__x~__ r8   r   )r   r   rL   r]   r   rT   r   r  r  r   r   r"  )rv  r   r   r   r   s   ``  @r6   "meta__adaptive_avg_pool2d_backwardr{  F  s   =D1d^^ 
 
MM!q f f f f f	
 	
 	
 	

 
L	TQYaaaa   
L
hn$_____   +M ,+>>$*%%((}(EEEr8   c                 b    t          | d           t          j        |t          j                  S )Nadaptive_avg_pool3d_backwardr   )!_adaptive_pool_empty_output_checkrL   r   r  r  r   s     r6   "meta__adaptive_avg_pool3d_backwardr  ]  s.     &k3QRRRD0NOOOOr8   r  c                       j         }t          d|          D ]3t          j                                       dk     fd           4d S )Nr   r   c                  $      dj          d dS )Nzc(): Expected grad_output to have non-zero size for non-batch dimensions, but grad_output has sizes rt  ru  r2  )r  r  r   s   r6   r[   z3_adaptive_pool_empty_output_check.<locals>.<lambda>i  s>     ` `-8->` `PQ` ` ` r8   )r   r   rL   r]   r   )r  r  r   r   s   `` @r6   r~  r~  d  s{    D1d^^ 
 
Q!#     	
 	
 	
 	

 
r8   c                 8     j         }t          j        |dv  fd           t          d|          D ]2t          j                                       dk     fd           3t          j        t          |          dk    d            d}d}d} j         dk    r                     d          }|dz  }                     |dz
            }|\  }} j         d	k    r?|||f}                     |          }	                     |t          j        
          }
|	|
fS ||||f}t          j	                   }                     |          
                    |          }	                     |t          j        
          
                    |          }
|	|
fS )Nr.   r  c                      d j          S )Nz:adaptive_max_pool2d(): Expected 3D or 4D tensor, but got: r2  r  s   r6   r[   z*meta_adaptive_max_pool2d.<locals>.<lambda>v      ZU[ZZ r8   r   r   c                       dj          d  dS )Nzjadaptive_max_pool2d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes rt  ru  r2  r   r   s   r6   r[   z*meta_adaptive_max_pool2d.<locals>.<lambda>{  4    T',{T TDET T T r8   r   c                      dS )NzCadaptive_max_pool2d(): internal error: output_size.size() must be 2r<   r<   r8   r6   r[   z*meta_adaptive_max_pool2d.<locals>.<lambda>      U r8   r  r.   r   r   )r   rL   r]   r   r   r   r   r   rF   r   r"  )r   rm  r   dimHsizeBsizeDosizeHosizeWr   r   r   r   r   s   `           @r6   meta_adaptive_max_pool2dr  p  s    :D	LZZZZ   1d^^ 
 
JJqMMA    	
 	
 	
 	
 
LKAUU  
 DEEzQ

1	JJtax  E NFFzQFF+	ooi((//)5;/??G|E662	3E::ooi((++-+HH//)5;/??BB' C 
 
 G|r8   c                 :     j         }t          j        |dv  fd           t           d           t          j        j         j        k     fd           t          j                  }                    j                  	                    |          S )Nr  c                      d j          S )NzKadaptive_max_pooling2d_backward(): Expected 3D or 4D grad_output, but got: r2  r  s   r6   r[   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s    q^i^oqq r8   adaptive_max_pool2d_backwardc                  (    dj          d j          S ry  r   )r  r   s   r6   r[   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s    c%+ccP[Pacc r8   r   )
r   rL   r]   r~  rT   rF   r   r   r   r"  )r  r   r   r   r   s   ``   r6   !meta_adaptive_max_pool2d_backwardr    s     D	Lqqqq  
 &k3QRRR	L{((ccccc  
 /66M??5;''***GGGr8   c                 :     j         }t          j        |dv  fd           t          d|          D ]2t          j                                       dk     fd           3t          j        t          |          dk    d            d}d}d}|dk    r                     d          }|dz  }                     |          }|\  }}}|d	k    r||||f}	n|||||f}	                     |	          }
                     |	t          j        
          }|
|fS )NrM  c                      d j          S )Nz:adaptive_max_pool3d(): Expected 4D or 5D tensor, but got: r2  r  s   r6   r[   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r  r8   r   r   c                       dj          d  dS )Nzjadaptive_max_pool3d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes rt  ru  r2  r  s   r6   r[   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r  r8   r.   c                      dS )NzCadaptive_max_pool3d(): internal error: output_size.size() must be 3r<   r<   r8   r6   r[   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r  r8   r!  r  r   )r   rL   r]   r   r   r   r   r   )r   rm  r   dimDr  r  osizeTr  r  r   r   r   r   s   `           @r6   meta_adaptive_max_pool3dr    sg    :D	LZZZZ   1d^^ 
 
JJqMMA    	
 	
 	
 	
 
LKAUU  
 DEEqyy

1	JJtE(FFFqyyFFF3		E666:	
//)
$
$Cooiu{o;;G<r8   c                 V    t          | d           |                    |j                  S )Nadaptive_max_pool3d_backward)r~  r   r   )r  r   r   s      r6   !meta_adaptive_max_pool3d_backwardr    s)     &k3QRRR??5;'''r8   c                 N    |t          d          |                     |          S )Nz:cannot repeat_interleave a meta tensor without output_size)r!  r   )repeatsrm  s     r6   meta_repeat_interleave_Tensorr    s+    WXXX[)))r8   c                 
   | j         j        sJ |j         j        sJ t          |                     t	          | j                             |                    t	          |j                             t
          j                  }|S Nr>   )rT   r   rJ   r"  r   r   rG   )realimagr  s      r6   meta_complexr    s{     :'''':''''+DJ7788+DJ77886>  F
 Mr8   )
fill_valuer  c                l    |                      ||                                 ft          j                  S r  )r   r{   rL   r   )r   r   r  s      r6   nonzero_staticr    s)     >>4,EJ>???r8   c                    t          j        t          j        d            t          j        |                                 |                                 fd|                                 ft           j        | j                  S )Nc                      dS )NaY  The register_meta function for torch.nonzero() raises unimplemented by default, as a correct data-independent implementation does not exist. This implementation returns a fake value, assuming all elements of the tensor are non-zero. To enable this registration, please set 'torch.fx.experimental._config.meta_nonzero_assume_all_nonzero' to True.r<   r<   r8   r6   r[   znonzero.<locals>.<lambda>  s
     S r8   r   rT   rx   )	rL   _check_not_implementedrW  meta_nonzero_assume_all_nonzeror!  r   r{   r   rx   r   s    r6   nonzeror    sy     
 2	S 	S   	txxzz"	
DJJLLj{	   r8   c           
      &	    t          j        t                    d            g }t                    D ]d\  Ft          j        j        t           j        t           j        t           j        t           j        fv d            j        t           j        t           j        fv rǉ                                }t          |          t          j
        j        z    j        k     fd           t          j                  D ]dt          j
        j                  j        z            k     fd           |                    |                    d                     e8|                               O|                               f|t          j        t                     j        k     fd           dd lm} t%           |j                   t                     j        k     r-                    d            t                     j        k     -d}d}D ]|dk    rd}|dk    rd	} nd
}|sg }g }t                    D ]1\  *|                               |                               2t                    D ]1\  *|                               |                               2                     |           |g g g t                    D ]_\  }	Dr!                     j        |	                    *                     j        |	                    Kt%          j                  `fd}
                     z   z             }ddlm}  |                                 dk              r|S  |
           }t3          j        |          }t%          |          t%          t          t          |                              k    r}t3          j        |j        |          }t3          j        |          }t3          j        |t3          j        |                    }|                    |                                |          }|S )Nc                      dS )Nz#at least one index must be providedr<   r<   r8   r6   r[   z#meta_index_Tensor.<locals>.<lambda>  s    (M r8   c                      dS )Nz?tensors used as indices must be long, int, byte or bool tensorsr<   r<   r8   r6   r[   z#meta_index_Tensor.<locals>.<lambda>  s    Y r8   c                      d j          S )N)too many indices for tensor of dimension r  r   s   r6   r[   z#meta_index_Tensor.<locals>.<lambda>%  s    S	SS r8   c            	      :    dj          d  dj          dz    S )NzThe shape of the mask 
 at index z0 does not match the shape of the indexed tensor r2  )r   r   jr]  r   s   r6   r[   z#meta_index_Tensor.<locals>.<lambda>*  sP     !h !h !hPQ !h !hJN*!h !h`ade`e!h !h r8   r   c                  :    dj          dt                      dS )Nr  z (got rn   )r   r   )r   r   s   r6   r[   z#meta_index_Tensor.<locals>.<lambda>5  s$    \DI\\SQX\\\\\ r8   r   Fr   Tc                    z   z   }t          |                                           }dgt                    z  |t                    t          | j                  t                    z
  <   |                     ||          S )zI
        This follows restride_src in TensorAdvancedIndexing.cpp
        r   )r   r   r   r   r>  )r   r   r   after_shapebefore_shapereplacement_shapes      r6   _restride_srcz(meta_index_Tensor.<locals>._restride_srcv  s     00;>t{{}}%%KL#PSQ
 Q
 K
L!!C
OOc+6F6F$FFG ug...r8   guard_or_false) rL   r]   r  	enumeraterT   r   r   r<  r  r   r   r   r   r   r   selecttorch._refs_refsr   r&   r   r   r  r  r   rF   3compute_elementwise_output_logical_to_physical_perm
apply_permr   invert_permr>  r   )r   r   r  r  refsstatehas_contiguous_subspacer  transposed_indicesr{   r  r   r  restrided_selfperm
perm_shaper/  r  r  r   r   r  r]  r  s   ``               @@@@@@@r6   meta_index_Tensorr    s   	Lg M MNNN &(Fg&& ! !5L
EIuz5:NNYY   {uz5:666--//KK"
Ndi/SSSS   uz** 8 8A&A$*QU*;;h h h h h h h h  
 MM'..A"6"677778 e$$$$MM%    G	LG	!\\\\\  
 (4('233G
g,,
"
"t g,,
"
" E# ' 'A:: aZZ}  ! #'
 # %!'** 	1 	1HAu A"))%000!'** 	1 	1HAu}A"))%000||D!!$ !LK#%(( 2 2
U=  5""4:c?3333##DJsO4444 $U[ 1 1	/ 	/ 	/ 	/ 	/ 	/ 	/ ..(99KG
H
HCDDDDDD~djjlla'(( 

 #]4((ND^TTD DzzT%D		**++++%ci66
6zBB
%j%2CD2I2IJJ
nnSXXZZ44Jr8   c                    d }d }d }|
d         r'|                      |                                          }|
d         r'|                      |                                          }|
d         r|                      |          }|||fS )Nr   r   r   r   r   )grad_output_input_weight_bias_sizes_optr   r  r  
transposedr  r  output_maskbackend_grad_inputbackend_grad_weightbackend_grad_biass                 r6   meta_convolution_backwardr    s      1~ C)33FKKMMBB1~ E*44W\\^^DD1~ C(22>BB 35FGGr8   c                                        d          }                     d          }|                     ||f          } t          j                                        dk    d            t          j                                        dk    d            t          j                             d                               d          k    fd           t          j                             d                               d          k    fd           t          j        |                      d          |k    o|                      d          |k    d	            |                     |                                            S )
Nr   r   r.   c                      dS rH  r<   r<   r8   r6   r[   zmeta_addbmm.<locals>.<lambda>  rI  r8   c                      dS rK  r<   r<   r8   r6   r[   zmeta_addbmm.<locals>.<lambda>  rI  r8   r   c                  `    d                      d           d                     d           S )Nz8batch1 and batch2 must have same number of batches, got r   r   r   rM  rN  s   r6   r[   zmeta_addbmm.<locals>.<lambda>  s3    p6;;WX>>pp`f`k`klm`n`npp r8   c            
          d                      d           d                      d           d                     d           d                     d           d	S )Nz#Incompatible matrix sizes for bmm (r   rC   r   r   rn   r   r  s   r6   r[   zmeta_addbmm.<locals>.<lambda>  sl    6&++a.. 6 66;;q>> 6 6;;q>>6 6$*KKNN6 6 6 r8   c                      dS )Nz.self tensor does not match matmul output shaper<   r<   r8   r6   r[   zmeta_addbmm.<locals>.<lambda>  s    @ r8   )r   r|  rL   r]   r{   r   )r   rM  rN  rI  rH  rY  rZ  s    ``    r6   meta_addbmmr    sg    ;;q>>D;;q>>D;;d|$$D	L"$H$HIII	L"$H$HIII	LA&++a..(ppppp   
LA&++a..(	
 	
 	
 	
 	
   
L		!51!5@@   >>$))++&&&r8   c                 P    |                      |                                           S r3   r  )r   r  kwargss      r6   meta_randint_liker    s    >>$))++&&&r8   )
grad_scale	found_infc       	         t    | |||||fD ]-t          j        t          t                    fd           .d S )Nc                  (    dt                      S Nz'exponent must be a tensor list but got ro   ls   r6   r[   z#meta__fused_adam_.<locals>.<lambda>      Gd1ggGG r8   rL   r]   rf   r   )r   gradsexp_avgsexp_avg_sqsmax_exp_avg_sqsstate_stepslrbeta1beta2weight_decayepsamsgradmaximizer  r  r  s                  @r6   meta__fused_adam_r     s]    & E8[/;O 
 
q$GGGG	
 	
 	
 	

 
r8   c       	             | |||||fD ]-t          j        t          t                    fd           .d } ||            ||           ||           ||           ||          fS )Nc                  (    dt                      S r  r  r  s   r6   r[   z"meta__fused_adam.<locals>.<lambda>  r  r8   c                     d | D             S )Nc                 6    g | ]}t          j        |          S r<   ri  )rB   r  s     r6   rE   z=meta__fused_adam.<locals>.empty_like_list.<locals>.<listcomp>  s#    999 ##999r8   r<   )tensor_lists    r6   empty_like_listz)meta__fused_adam.<locals>.empty_like_list  s    99[9999r8   r  )r   r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  s                   @r6   meta__fused_adamr    s    & E8[/;O 
 
q$GGGG	
 	
 	
 	

: : : 	!!$$(( r8   c                 r    t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        u  fd           t          j        j        t           j        u fd           t          j                             d                              d          k     fd                                                     d                              d          ft           j        	          S )
Nr   c                      dS )Nza must be a 2D tensorr<   r<   r8   r6   r[   zmeta__int_mm.<locals>.<lambda>      '> r8   c                      dS )Nzb must be a 2D tensorr<   r<   r8   r6   r[   zmeta__int_mm.<locals>.<lambda>  r
  r8   c                      d j          S )Nzexpected self to be int8, got r   )r  s   r6   r[   zmeta__int_mm.<locals>.<lambda>      ::: r8   c                      d j          S )Nzexpected mat2 to be int8, got r   )r  s   r6   r[   zmeta__int_mm.<locals>.<lambda>  r  r8   r   r   c            
          d                      d           d                      d           d                     d           d                     d           d	S )Nz'Incompatible matrix sizes for _int_mm (r   rC   r   r   rn   r   r  r  s   r6   r[   zmeta__int_mm.<locals>.<lambda>"  sl    ,affQii , ,!&&)) , ,66!99, , vvayy, , , r8   r   )rL   r]   r{   rT   r<  r   r   r=  r  s   ``r6   meta__int_mmr    s"    
LA>>???	LA>>???	L	5:::::   
L	5:::::   
L	q		QVVAYY	
 	
 	
 	
 	
   ;;q		166!99-U[;AAAr8   c                 j    t          j                                         dk    d            t          j         j        t           j        u  fd                                d          }                     d          dz  }                     |dz  ||dz  z  d|dz  ft           j        	          S )
Nr   c                      dS Nzw must be a 2D tensorr<   r<   r8   r6   r[   z2meta__convert_weight_to_int4pack.<locals>.<lambda>,  r
  r8   c                      d j          S Nr  r   r  s   r6   r[   z2meta__convert_weight_to_int4pack.<locals>.<lambda>/      8qw88 r8   r   r      rU      r   )rL   r]   r{   rT   r  r   r   r=  r  inner_k_tilesr  r]  s   `   r6    meta__convert_weight_to_int4packr  *  s    	LA>>???	L	5;8888   	
q		A	q		AA;;F-"$%Q		
 k    r8   c                 N    t          j                                         dk    d            t          j         j        t           j        u  fd                                d          }                     d          }                     ||dz  ft           j                  S )Nr   c                      dS r  r<   r<   r8   r6   r[   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>@  r
  r8   c                      d j          S Nzexpected w to be int32, got r   r  s   r6   r[   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>C  r  r8   r   r   r   )rL   r]   r{   rT   r=  r   r   r  r  s   `   r6   (meta__convert_weight_to_int4pack_for_cpur"  >  s    	LA>>???	L	5;8888   	
q		A	q		A;;	
AFk    r8   c                     t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j        j        t           j        u fd                                 	                    d          	                    d          dz   j        	          S )
Nr   c                      dS Nzx must be a 2D tensorr<   r<   r8   r6   r[   z*meta__weight_int4pack_mm.<locals>.<lambda>O  r
  r8   r  c                      dS )Nzw must be a 4D tensorr<   r<   r8   r6   r[   z*meta__weight_int4pack_mm.<locals>.<lambda>P  r
  r8   c                      d j          S Nr  r   rC   s   r6   r[   z*meta__weight_int4pack_mm.<locals>.<lambda>S      ?ag?? r8   c                      d j          S r!  r   r  s   r6   r[   z*meta__weight_int4pack_mm.<locals>.<lambda>W  r  r8   r   r  r   
rL   r]   r{   rT   rV  rW  rX  r=  r   r   r  s   ``  r6   meta__weight_int4pack_mmr-  M  s    	LA>>???	LA>>???	L	EM5=%.AA????   
L	5;8888   ;;qvvayy!&&))a-qw;???r8   c                     t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j        j        t           j        u fd                                 	                    d          	                    d           j                  S )Nr   c                      dS r%  r<   r<   r8   r6   r[   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>^  r
  r8   c                      dS r  r<   r<   r8   r6   r[   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>_  r
  r8   c                      d j          S r(  r   r)  s   r6   r[   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>b  r*  r8   c                      d j          S r  r   r  s   r6   r[   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>f  r  r8   r   r   )
rL   r]   r{   rT   rV  rW  rX  r  r   r   r  s   ``  r6    meta__weight_int4pack_mm_for_cpur3  \      	LA>>???	LA>>???	L	EM5=%.AA????   
L	5;8888   ;;qvvayy!&&))17;;;;r8   c                     t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j        j        t           j        u fd                                 	                    d          	                    d           j                  S )Nr   c                      dS r%  r<   r<   r8   r6   r[   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>m  r
  r8   c                      dS r  r<   r<   r8   r6   r[   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>n  r
  r8   c                      d j          S r(  r   r)  s   r6   r[   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>q  r*  r8   c                      d j          S r!  r   r  s   r6   r[   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>u  r  r8   r   r   r,  )rC   r  r  qScaleqZeross   ``   r6   )_weight_int4pack_mm_with_scales_and_zerosr<  k  r4  r8   r  r  c                     | |z   dz
  |z  |z  S r*  r<   r  s     r6   kai_roundupr>  z  s    UQY1!!r8   c                 0  	
 | dk    r||k    r+d}d}d}dddd fdfd} ||||||          S |dz  d	k    rH||z  d	k    rAd}d}d}dddd		fd
}	
fdd 
	fd	fd |||||||          S d S d S d S )Nr  r  rU  r   c                 H    t          ||z  d          }t          | |          S )Nr  r>  )r]  krsrkr_sr_roundedup4s       r6   kai_k_roundedupz3get_kai_packed_weight_size.<locals>.kai_k_roundedup  s)     $/rBw#:#: "1&6777r8   c                 f     | ||          }|dz  dk    s
J d            ||dz  z   z   z   z  S )Nr   r   zk_internal must be evenr<   )	r]  nrrB  rC  
k_internalrE  kai_num_bytes_biaskai_num_bytes_multiplier_rhskai_num_bytes_sum_rhss	        r6   9kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0z]get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0  sd     -_QB77
"Q1,,,.G,,,1_23+, )) r8   c                 L    t          | |          |z  }| ||||          z  S r3   rA  )r  r]  rG  rB  rC  num_rowsrL  s         r6   7kai_get_rhs_packed_size_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0z[get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_size_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0  sA     'q"--3 OO2r2 r8   r  r   c                     ||z  dk    sJ |	z  dk    sJ |z  dk    sJ t          | |          |z  }| |||||          z  S rk  rA  )
r  r]  rG  rB  rC  blrN  kai_bl_multiple_of;kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0kai_nr_multiple_ofs
          r6   9kai_get_rhs_packed_size_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0z]get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_size_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0  s     RA~~~~//A5555//A5555&q"--3 QQ2r2r r8   c                     ||z  dk    sJ |
z  dk    sJ |z  dk    sJ  	            } | |          } ||          }|||z  z   z   z  S rk  r<   )r]  rG  rB  rC  rQ  num_bytes_multiplier_rhsnum_blocks_per_rownum_bytes_per_blockrR  #kai_get_bf16_datatype_size_in_bytesrT  kai_num_blocks_per_rowrI  kai_num_bytes_per_blockrK  s           r6   rS  z_get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0  s     RA~~~~//A5555//A5555 ,O+N+P+P(%;%;Ar%B%B"&=&=0' '# (+==+,() r8   c                      dS )Nr   r<   r<   r8   r6   rZ  zGget_kai_packed_weight_size.<locals>.kai_get_bf16_datatype_size_in_bytes  s    qr8   c                 @    |z  dk    sJ t          | |          |z  S rk  rA  )r]  rQ  rR  s     r6   r[  z:get_kai_packed_weight_size.<locals>.kai_num_blocks_per_row  s/    //A5555"1b))R//r8   c                 *    | z  dk    sJ | dz  |z   S )Nr   r   r<   )rQ  rW  rR  s     r6   r\  z;get_kai_packed_weight_size.<locals>.kai_num_bytes_per_block  s*    //A5555a#;;;r8   r<   )n_bitsr  K	groupsizekai_nrkai_krkai_srrO  rU  rR  rZ  rS  rL  rE  rT  r[  rI  rJ  r\  rK  s            @@@@@@@@@@@r6   get_kai_packed_weight_sizerf  ~  s   {{>>FFF$%!+,(!"8 8 8       
 
 
 
 
 KJ1fff   ^q  Q]a%7%7FFF$%!!"!"!#                 ,  0 0 0 0 0< < < < < ML1fffi  O {Z ! %7%7r8   c                 @    t          j         j        t           j        u  fd           t           j        j                                        r||k    r|j        t           j        k    s-||k     rg|dz  dk    r^||z  dk    rU|j        t           j        k    r@t          d|||          } 
                    t          |          t           j                  S                                  |                                z   } 
                    |t           j                  S )Nc                      d j          S r  r   )weightss   r6   r[   z2meta__dyn_quant_pack_4bit_weight.<locals>.<lambda>  s    >w}>> r8   r  r   r  r   )rL   r]   rT   r  backendskleidiaiis_availablerP   rX  rf  r   r   r   )ri  scales_zerosr7  
block_sizein_featuresout_featurespacked_weight_sizes   `      r6    meta__dyn_quant_pack_4bit_weightrr    s    
L$>>>>   ~++-- M	{	"	"|'9U['H'H$$R1$$j(A--"en44 8|[*
 
   %7!8!8 LLL <+=+=+?+??/u{CCCr8   c                     t          j                                         dk    d            t          j         j        t           j        fv  fd                                d          }                     || j                  S )Nr   c                      dS )Nzinput must be a 2D tensorr<   r<   r8   r6   r[   z-meta__dyn_quant_matmul_4bit.<locals>.<lambda>  s    )D r8   c                      d j          S )Nzexpected input to be f32, got r   )inps   r6   r[   z-meta__dyn_quant_matmul_4bit.<locals>.<lambda>  s    <<< r8   r   r   )rL   r]   r{   rT   rV  r   r   )rv  packed_weightsrn  ro  rp  r  s   `     r6   meta__dyn_quant_matmul_4bitrx    s     
La!D!DEEE	L	em_$<<<<   	A==L	=:::r8   c                     t          j                                         dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j                                        dk    d            t          j        j        t           j        u fd                                 	                    d          	                    d           j                  S )Nr   c                      dS r%  r<   r<   r8   r6   r[   z*meta__weight_int8pack_mm.<locals>.<lambda>  r
  r8   c                      d j          S r(  r   r)  s   r6   r[   z*meta__weight_int8pack_mm.<locals>.<lambda>  r*  r8   c                      dS r  r<   r<   r8   r6   r[   z*meta__weight_int8pack_mm.<locals>.<lambda>  r
  r8   c                      d j          S )Nzexpected w to be int8, got r   r  s   r6   r[   z*meta__weight_int8pack_mm.<locals>.<lambda>  s    7ag77 r8   r   r   )
rL   r]   r{   rT   rV  rW  rX  r<  r   r   )rC   r  q_scaless   `` r6   meta__weight_int8pack_mmr    s    	LA>>???	L	EM5=%.AA????   
LA>>???	L	5:7777   ;;qvvayy!&&))17;;;;r8   c                     t          j                                         dk     fd           t          j                                        dk    fd           t          j                             d                              d          k     fd           t          j        t	          j         j                  d            t          j        t	          j        j                  d            t          j        |dk    d	            t          j        d
v fd                                d          }                    d          } j        d d         }j        d d         }t          t          j	        ||                    }|
                    ||g                                |          S )Nr   c                  4    d                                   dS )Nz1cdist only supports at least 2D tensors, X1 got: r  r   )x1s   r6   r[   z$meta_cdist_forward.<locals>.<lambda>(      OBFFHHOOO r8   c                  4    d                                   dS )Nz1cdist only supports at least 2D tensors, X2 got: r  r   )x2s   r6   r[   z$meta_cdist_forward.<locals>.<lambda>,  r  r8   r   c                  `    d                      d           d                     d           S )Nz4X1 and X2 must have the same number of columns. X1: r   z X2: r   )r  r  s   r6   r[   z$meta_cdist_forward.<locals>.<lambda>0  s2    frwwr{{ffY[Y`Y`acYdYdff r8   c                      dS )Nz=cdist only supports floating-point dtypes, X1 got: {x1.dtype}r<   r<   r8   r6   r[   z$meta_cdist_forward.<locals>.<lambda>4  r   r8   c                      dS )Nz=cdist only supports floating-point dtypes, X2 got: {x2.dtype}r<   r<   r8   r6   r[   z$meta_cdist_forward.<locals>.<lambda>8  r   r8   r   c                      dS )Nz)cdist only supports non-negative p valuesr<   r<   r8   r6   r[   z$meta_cdist_forward.<locals>.<lambda>:  s    !L r8   Nr   r   c                      d  S )Nz%possible modes: None, 1, 2, but was: r<   )compute_modes   r6   r[   z$meta_cdist_forward.<locals>.<lambda>=  s    FFF r8   r  )rL   r]   r{   r   rF   is_float_dtyperT   r   r   broadcast_shapesextendr   )	r  r  r:  r  r1r2batch_tensor1batch_tensor2r_  s	   `` `     r6   meta_cdist_forwardr  $  s   	L
AOOOO   
L
AOOOO   
L
rwwr{{"fffff   
LRX&&OO   
LRX&&OO   
LaLLMMM	L$FFFF   
B	BHSbSMMHSbSMM.}mLLMMLR!!!<<%%%r8   c                 @   |j         d         }|j         d         }|j         d         }|j         d d         }|j         d d         }	t          t          j        ||	                    }
|
                                }|                    ||g           t          j        |
          }|dk    s|dk    s|dk    s|dk    rt          j        |          S |t          |j                   k    r|	                    |          }t          j
        |t          j                  S )Nr   r  r   r   )r   r   rL   r  copyr  mathprod
zeros_liker|  r   r   )r>  r  r  r:  cdistc1r  r  r  r  rx  tensor1_expand_sizebatch_products                r6   meta_cdist_backwardr  H  s
    
"B	"B	"BHSbSMMHSbSMM 6}m T TUU.3355Bx(((I233M	Qww"''R1WW(:(:###d28nn,,YY*++Be.EFFFFr8   c	                 r    t          j        j        t           j        t           j        fv fd           t          j        j        t           j        t           j        fv fd           t          j        t          j         j                   fd                               d          }	|rt          j        |	dk    d            |	dz  }	                     |	                     d                    }
t          j        |t          k    d            t          j        j
        dk    fd           t          j                                                                        k    fd	           fd
d fd}t                    dk    r                                        d                    }                                                              }|t          k    r+                    |	                     d                    }n                    d          }n | |
|          }|t          t          fv s|s)                                        d                    }n                    d          }                    |	          }j        d         }|t          k    rC|rt          j        |dk    d            |dz  }                    | j        d                   }n'                    |                                          }|
|||fS )Nc                      d j          S )Nz(expected indices to be long or int, got r   )r   s   r6   r[   z$meta_embedding_bag.<locals>.<lambda>m      J7=JJ r8   c                      d j          S )Nz(expected offsets to be long or int, got r   )ro  s   r6   r[   z$meta_embedding_bag.<locals>.<lambda>q  r  r8   c                      d j          S )Nz/expected weight to be floating point type, got r   )r5  s   r6   r[   z$meta_embedding_bag.<locals>.<lambda>u  s    P&,PP r8   r   r   c                      dS Nz1include_last_offset: numBags should be at least 1r<   r<   r8   r6   r[   z$meta_embedding_bag.<locals>.<lambda>|  s    G r8   c                      dS )Nz@embedding_bag: per_sample_weights only supported with mode='sum'r<   r<   r8   r6   r[   z$meta_embedding_bag.<locals>.<lambda>  s    V r8   c                      d j          dS )Nz1expected per_sample_weights to be 1D tensor, got r  r  )per_sample_weightss   r6   r[   z$meta_embedding_bag.<locals>.<lambda>  s    bHZH_bbb r8   c                  ^    d                                  d                                   dS )Nz%expected per_sample_weights.numel() (z$ to be the same as indices.numel() (rn   r   )r   r  s   r6   r[   z$meta_embedding_bag.<locals>.<lambda>  sF    I8J8P8P8R8R I I6=mmooI I I r8   c                 P     | ||          o|                     d          dk    S Nr   r   r   )r%  rs  r   padding_idxis_fast_path_index_selects       r6   is_fast_path_index_select_scalez;meta_embedding_bag.<locals>.is_fast_path_index_select_scale  s0    %%c6;??XELLQROOWXDX	
r8   c                     | j         t          j        k    s| j         t          j        k    o7|                     d          dk    o|                    d          dk    o|dk     S Nr   r   )rT   rL   rP   rN   r   )r%  r   r  s      r6   r  z5meta_embedding_bag.<locals>.is_fast_path_index_select  s_    Y%+%@ej)@  

1" a  A%  a		
r8   c                 >    | | |||          S  | ||          S r3   r<   )r%  rs  r   r  r  r  s       r6   is_fast_pathz(meta_embedding_bag.<locals>.is_fast_path  s5    223v{SSS,,S&+FFFr8   cpuc                      dS r  r<   r<   r8   r6   r[   z$meta_embedding_bag.<locals>.<lambda>  s    O r8   )rL   r]   rT   r   r   rF   r  r   r   MODE_SUMr   r   r   MODE_MAX	MODE_MEANr   )r5  r   ro  scale_grad_by_freqrU  sparser  include_last_offsetr  num_bagsr   r  
offset2bagbag_sizemax_indicesfast_path_sumnumBagsr  r  s   ```   `          @@r6   meta_embedding_bagr  _  s    
L%*ei00JJJJ   
L%*ei00JJJJ   
LV\**PPPP  
 ||AH MGG	
 	
 	
 	AhA77F%HVV	
 	
 	
 	#q(bbbb	
 	
 	
 	$$&&'--//9    	
 	
 	

 
 
 
 


 
 
G G G G G G 7u$$&&w||A77
$$W\\^^448!++Hfkk!nnEEKK!++A..KK$V-?UUIx(((( **7<<??;;JJ **1--J$$X..-"8" qLOO   1!++GV\!_EEKK!++HMMOO<<K:x44r8   c                     t          | ||g|R  \  }}}}t          |          dk    r'|                    |                                          }||||fS )Nr  )r  r   r   r   )r5  r   ro  rH   r   r  r  r  s           r6   meta_embedding_bag_forward_onlyr    sm    0B1#'1 1 1-FJ+ 7u$$$$W\\^^44:x44r8   c                 r    |r|S | j         j        s| j         j        r| j         S |rt          j        S | j         S r3   )rT   r   r   rL   r   )r   rT   promote_int_to_longs      r6   _get_reduction_dtyper    sH     {$ (> {	 z;r8   r   c                    t          | |d          }t          j        | j        |          }t	          | ||          }|                     ||          S )NT)r  r   )r  rF   r  r   r  r   )r   r  r  rT   r  r_  s         r6   meta_nansumr    sS     (u$OOOLT22D+E4AAL??<|?<<<r8   c           	          t          j        | j        t          t	          |                                                               }|                     |          S r3   )rF   r  r   r\   r   r{   r   )r   r_  s     r6   meta_medianr    sG    7U5--.. L ??<(((r8   c                    t          |           dk    rt          j        d           t          j        | j        |f          }t          | ||          }|                     |          |                     |t          j                  fS )Nr   zmedian CUDA with indices outputr   )	r   rF   alert_not_deterministicr  r   r  r   rL   r   )r   r{   r  r_  s       r6   meta_median_mode_dimr    s~     5V##%&GHHH

u{SF
3
3C+E3@@L%%EJ77 r8   c                     | S r3   r<   r   s    r6   meta_logical_not_r    re  r8   c                    t          j        t                    |                                 k    d            t	                    D ]"\  t          j        dk    fd           #t                    |                                 z
  }d|z  t          | j                  z   fdt          t                              D             }|                     |          S )Nc                      dS )NzZNumber of dimensions of repeat dims can not be smaller than number of dimensions of tensorr<   r<   r8   r6   r[   zmeta_repeat.<locals>.<lambda>
  s    l r8   r   c                      d d  S )Nz"Repeats cannot be negative, found r  r<   )r   reps   r6   r[   zmeta_repeat.<locals>.<lambda>  s    KKKKK r8   r  c                 2    g | ]}|         |         z  S r<   r<   )rB   r   padded_sizer  s     r6   rE   zmeta_repeat.<locals>.<listcomp>  s&    LLL1;q>GAJ.LLLr8   )	rL   r]   r   r{   r  r\   r   r   r   )r   r  num_new_dimensionstarget_sizer   r  r  s    `  @@@r6   meta_repeatr    s    	LG

"ll   G$$ 
 
31HKKKKK	
 	
 	
 	
 W

2++eDJ.?.??KLLLLLc'll8K8KLLLK>>+&&&r8   c                     | S r3   r<   r   s    r6   
meta_zero_r    re  r8   c                 n    t          |t          j                  rt          | j        |j                   | S r3   )rf   rL   r   r_   r   r   r   s     r6   meta_binop_inplacer    s1     %&& 9
EK888Kr8   c                    d }d }d } ||           r ||          rt          d           ||           r ||          st          d          t          |t          j                  rt	          | j        |j                   | S )a*  
    Some checks for inplace ops.
    Checks for promotion rules for some dtypes.
    int.add/sub_(float) and bool.add/sub_(others) are rejected.
    Promoting in these in-place operations would require reallocating
    and copying over elements, hence not allowed.
    Checks for alpha param.
    c                     t          | t                    rt          j        | j                  S t          | t
                    S r3   )rf   r   rF   r7  rT   r   rh   s    r6   is_integericz.meta_binop_inplace_alpha.<locals>.is_integericB  s7    c:&& 	,)#)444c7+++r8   c                     t          | t                    rt          j        | j                  S t          | t
                    S r3   )rf   r   rF   r  rT   r   r  s    r6   
is_floaticz,meta_binop_inplace_alpha.<locals>.is_floaticH  s7    c:&& 	.'	222c9---r8   c                     t          | t                    rt          j        | j                  S t          | t
                    S r3   )rf   r   rF   is_boolean_dtyperT   r   r  s    r6   is_booleanicz.meta_binop_inplace_alpha.<locals>.is_booleanicN  s7    c:&& 	-)#)444c8,,,r8   z]Promotion of int.add/sub_(float) in in-place ops are not possible due to element size change.z_Promotion of book.add/sub_(others) in in-place ops are not possible due to element size change.)r!  rf   rL   r   r_   r   )r   r   rH  r  r  r  s         r6   meta_binop_inplace_alphar  0  s    $, , ,. . .- - - |D 
jj// 
k
 
 	

 |D 
,,u"5"5 
m
 
 	
 %&& 9
EK888Kr8   c                 :    t          | |t          j                  S r  rJ   r   rG   r   r   rH  s      r6   meta_binop_alphar  e  s&     e$C$K   r8   c                 8    t          | t          j                  S r  r  )r   r  s     r6   
meta_roundr  q  s"    <D   r8   c                 \    t          j        t          j        j                   fd           t          t           j                  r2t          j        t          j        j                   fd           d S t          j        t          t                     fd           d S )Nc                        dj          S )Nz7: Expected input tensor to have an integral dtype. Got r   )r  r   s   r6   r[   z#shift_dtype_check.<locals>.<lambda>{  s    7__SWS]__ r8   c                        dj          S )Nz6: Expected shift value to have an integral dtype. Got r   r  r  s   r6   r[   z#shift_dtype_check.<locals>.<lambda>  s    waaVYV_aa r8   c                        d S )Nz): Expected shift value to be an int. Got r<   r  s   r6   r[   z#shift_dtype_check.<locals>.<lambda>  s    wNNNN r8   )rL   r]   rF   r7  rT   rf   r   r   )r  r   r  s   ```r6   shift_dtype_checkr  x  s    	Ltz**_____   #u|$$ 	
"39--aaaaa	
 	
 	
 	
 	

 	sG$$NNNNN	
 	
 	
 	
 	
r8   c                 \    t          d| |           t          | |t          j                  S )Nrshiftr  r  rJ   r   rG   r  s     r6   meta_rshiftsr    6    he,,,e$C$K   r8   c                 \    t          d| |           t          | |t          j                  S )Nlshiftr  r  r  s     r6   meta_lshiftsr    r  r8   c                 6    |                      | j                  S r3   r  r   s    r6   	meta_zeror    s    >>$*%%%r8   c                     | S r3   r<   r   r  s     r6   
meta_fill_r    re  r8   c                 *    t          j        |           S r3   ri  r  s     r6   	meta_fillr        D!!!r8   c                     | S r3   r<   r   s    r6   
meta_relu_r    re  r8   c                 :    t          | |t          j                  S r  r  r  s      r6   meta__add_relur	    s&     e$C$K   r8         ?UUUUUU?c                 *    t          j        |           S r3   ri  r   noiselowerr  r  r   s         r6   meta_rrelu_with_noiser    s    
 D!!!r8   c                 R    t          j        |           t          j        |          fS r3   ri  r  s         r6    meta_rrelu_with_noise_functionalr    s%     D!!5#3E#:#:::r8   c                     | S r3   r<   )r   r  r  r  r   s        r6   meta_rrelu_with_noise_r    s	     Kr8   c                 *    t          j        |           S r3   ri  r   r   r   
accumulates       r6   meta_index_putr    r  r8   c                 :    t          | j        |j                   | S r3   r_   r   )r   r{  values      r6   meta_masked_fill_r    s    DJ
333Kr8   c                     |                      |                                                               t          j        |                     }|S r   )r   r   r"  rF   r   )r   r{  rs  masked_scales       r6   meta__masked_scaler    sF    >>$))++..111$77 2  L r8   c                      t          j        |j        t           j        t           j        fv d            t          j         j        j        k     fd            S )Nc                      dS )NzMask must be bool or uint8r<   r<   r8   r6   r[   z&meta_masked_scatter_.<locals>.<lambda>  s    9U r8   c                  (    d j          dj          S )NzEmasked_scatter: expected self and source to have same dtypes but got r   r   )r   rb  s   r6   r[   z&meta_masked_scatter_.<locals>.<lambda>  s(     :*: :+1<: : r8   )rL   r]   rT   r  r  )r   r{  rb  s   ` `r6   meta_masked_scatter_r#    ss    	L
uz5;//1U1U   
L
fl"	: 	: 	: 	: 	:  
 Kr8   c                     t          | |          \  } }t          j        | t          j                  }t	          |||          S r   )r&   rL   r   r   r#  )r   r{  rb  r   s       r6   meta_masked_scatterr%    sA     "$--JD$d%2IJJJFf555r8   c                 ,    |                      |          S r3   r}  )r   r{  r@  s      r6   meta_masked_scatter_backwardr'    s    >>%   r8   c                     | S r3   r<   r  s       r6   meta_index_put_r)    re  r8   c                   
 t          j        |                                 dk    d            t          j        |                                dk    d            |                                 }|                                
|d         |d         |d         }
d         }||ft          j        
d         k    o
d         k    
fd           |r| j        t           j        k    p| j        t           j        k    o|t           j        k    }t          j        || j        k    p|d            |                              	                    |          }	n|                              }	|s][t          j                                        dk    d	            t          j                                        k    fd
           |	S )Nr.   c                      dS rH  r<   r<   r8   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>  rI  r8   c                      dS rK  r<   r<   r8   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>  rI  r8   r   r   r   c            	      :    d d d d          d d          d	S rP  r<   rR  s   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>  sU     RSU R RR R+7?R R>J1oR R R r8   c                      dS )Nzfout_dtype only supported for torch.float32 output with float16/bfloat16 inputs or same as input dtypesr<   r<   r8   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>  s    | r8   c                      dS )Nzself must be a 3D tensorr<   r<   r8   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>  s    6P r8   c                  8    d  d                                  S )Nz*Expected an input tensor shape with shape z but got shape: r   )rm  self_baddbmms   r6   r[   z)common_meta_baddbmm_bmm.<locals>.<lambda>   s%    sss^j^o^o^q^qss r8   )
rL   r]   r{   r   rT   rW  rX  rV  r   r"  )rM  rN  is_bmmr1  r9  r\  res_rowsres_colssupported_out_dtyper   rS  rT  rU  rm  s      `      @@@@r6   common_meta_baddbmm_bmmr6    s   	L"$H$HIII	L"$H$HIII;;==L;;==L	aB#AAHAHx*K	LQ2E,q/5E"E	R 	R 	R 	R 	R 	R  
  /LEM)KV\U^-K)5=( 	 	%<)<||	
 	
 	
 !!+..11)<< !!+.. 
l.\%%''1,.P.PQQQ;.sssss	
 	
 	

 Mr8   c                 $    t          | |d          S )NTr6  )r   rD  s     r6   meta_bmmr9  &  s    "4t444r8   c                 (    t          | |d|          S )NT)r9  r8  )r   rD  r9  s      r6   meta_bmm_dtyper;  +  s    "4tyIIIIr8   c                     | |z  }| |z  }|dk    r-t          |dk               t          |dk               k    r|dz  }|S r  )r  )rC   yqrY  s       r6   div_rtnr?  0  sK    	QA	AA 	Avv4A;;$q1u++--	QHr8   c                     t          | |z   |z   ||dz
  z  z
  dz
  |r|dz
  ndz   |          dz   }|r|dz
  |z  | |z   k    r|dz  }|S r  )r?  )	inputSize
kernelSizer  r  r   r  r  
outputSizes           r6   pooling_output_shape_pad_lrrD  :  s     	 *q.)* 	
 '-vzzA/ 	
 	
 		   Nf$	E(999!OJr8   c           	          t          j        |dk    d            t          j        dk    fd           t          j        dz
  z  dz   dz  k    fd           t          | ||          S )Nr   c                      dS )Nzstride should not be zeror<   r<   r8   r6   r[   z&pooling_output_shape.<locals>.<lambda>V  s    &A r8   c                      d  S )Nz'pad must be non-negative, but got pad: r<   pads   r6   r[   z&pooling_output_shape.<locals>.<lambda>W  s    #RS#R#R r8   r   r   c                      d d d  S )NzApad should be at most half of effective kernel size, but got pad=z, kernel_size=z and dilation=r<   )r  rB  rI  s   r6   r[   z&pooling_output_shape.<locals>.<lambda>Z  s;    @PS @ @%@ @5=@ @ r8   )rL   r]   rD  )rA  rB  rI  r   r  r  s    `` ` r6   r(  r(  U  s    	L1AABBB	LRRRRSSS	La8+a/A55	
 	
 	
 	
 	
 	
   ':sC9  r8   c           	      L   	
                                   }	t          j        dk    odk    d            t          j        |dk    o|dk    d            t          j        |dk    o|dk    d                                 d          dk    o                     d          dk    }|t          j        k    r6t          j        |dk    o|o                     d          dk    d	            nXt          j        |dk    r                     d          dk    r|p |dk    o|o                     d          dk     fd
           t          j        dz  k    odz  k    fd           t          j        dk    odk    
	fd           d S )Nr   c                      dS )NzCkernel size should be greater than zero, but got kH: {kH}, kW: {kW}r<   r<   r8   r6   r[   z$pool2d_shape_check.<locals>.<lambda>z  r  r8   c                      dS )Nz>stride should be greater than zero, but got dH: {dH}, dW: {dW}r<   r<   r8   r6   r[   z$pool2d_shape_check.<locals>.<lambda>~  s    P r8   c                      dS )Nz\dilation should be greater than zero, but got dilationH: {dilationH}, dilationW: {dilationW}r<   r<   r8   r6   r[   z$pool2d_shape_check.<locals>.<lambda>  s    n r8   r   r   r  r.   c                      dS )NzExpected 4D (batch mode) tensor expected for input with channels_last layout with optional 0 dim batch size for input, but got: {input.size()}r<   r<   r8   r6   r[   z$pool2d_shape_check.<locals>.<lambda>  s
     Q r8   c                  2    d                                   S )NzYExpected 3D or 4D (batch mode) tensor with optional 0 dim batch size for input, but got: r   r  s   r6   r[   z$pool2d_shape_check.<locals>.<lambda>  s    ~pupzpzp|p|~~ r8   c                       d d d d  S )NzKpad should be smaller than or equal to half of kernel size, but got padW = z	, padH = z, kW = z, kH = r<   )r,  r-  r0  r1  s   r6   r[   z$pool2d_shape_check.<locals>.<lambda>  s>     >> >!%> >.0> >9;> > r8   c                  .    d d  d d d d dS NzGiven input size: (rC   z). Calculated output size: (z). Output size is too smallr<   )r2  r3  r  r8  r  r  s   r6   r[   z$pool2d_shape_check.<locals>.<lambda>  sZ     #k # #K # #* # #$0# #3?# #BM# # # r8   )r{   rL   r]   r   r  )r   r,  r-  r.  r/  r0  r1  	dilationH	dilationWr  r2  r3  r  r  r   r   
valid_dimsr8  s   ```  ``  `````   @r6   r)  r)  d  s8   " 99;;DL	L
Q26UU   
L
Q26PP   
LA')a-nn  
 A!#:

1(:J+++AI;*;A!);Q Q	
 	
 	
 	
 	QY<5::a==A-<* A	?j?UZZ]]a-?~~~~	
 	
 	
 
L
a4+B!GtO	> 	> 	> 	> 	> 	> 	>   
Lq.\Q.	# 	# 	# 	# 	# 	# 	# 	# 	#    r8   rW  rT  r,  r-  rU  r.  r/  pTpHpW	dilationTrT  rU  rX  rY  rZ  r[  r\  r]  rR  c           
      8   	
  j         }t          j        dk    odk    odk    fd           t          j        dk    odk    odk    fd           t          j        dk    odk    odk    fd           t          j        |dv  fd           t          |          D ]@|dk    rdk    rt          j                                       dk     fd           A|r-t          j        k    ok    ok    fd	           t          j        d
z  k    od
z  
k    od
z  	k    	
fd           t          j        dk    odk    odk    fd           d S )Nr   c                      d d  d S )Nz5kernel size should be greater than zero, but got kT: z, kH: z, kW: r<   )r,  rT  r-  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  s4    ,, ,, ,'), , r8   c                      d d  d S )Nz0stride should be greater than zero, but got dT: z, dH: z, dW: r<   )r.  rU  r/  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  s$    WrWWWWSUWW r8   c                      d d  d S )Nz9dilation should be greater than zero, but got dilationT: z, dilationH: z, dilationW: r<   )rT  rZ  rU  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  s;    V#V V2;V VJSV V r8   rM  c                        dj          S )Nz/: Expected 4D or 5D tensor for input, but got: r2  )r  r   s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  s    7XX5;XX r8   r!  c                  J      dj          d                               dS )NzZ: Expected input's non-batch dimensions to have positive length, but input has a shape of z and non-batch dimension z has length zero!)r   r   )r  r   r   s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  sG     M M-2[M M,1JJqMMM M M r8   c                  .    d d  d d d d dS )Nzinput image (T: r)  r  z ) smaller than kernel size (kT:  kH:  kW: rn   r<   )rY  rX  rZ  r,  rT  r-  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  s]    <5 < <g < <6 < <$&< <-/< <68< < < r8   r   c                  ,    d d d  d d d S )NzHpad should be smaller than or equal to half of kernel size, but got kT: rc  rb  z padT: z padW: z padH: r<   )r,  rT  r-  rX  rW  rY  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  sh    KK KK K%'K K02K K;=K KFHK K r8   r   c                  :    d d d  d d d d d dS rS  r<   )rY  rX  rZ  rW  r\  r[  r]  s   r6   r[   z$pool3d_shape_check.<locals>.<lambda>  sv    (' ( (E ( (G ( (f ( ((/( (27( (:A( (DJ( ( ( r8   )r   rL   r]   r   r   )r   rW  rT  r,  r-  rU  r.  r/  rW  rX  rY  rZ  rT  rU  rX  rY  rZ  r[  r\  r]  r  rR  r   r   s   `````````````````````  @r6   rS  rS    s   0 :D	L
Q$26$b1f	
 	
 	
 	
 	
 	
   
L
Q$26$b1f	
 	
 	
 	
 	
 	
   
LA9)a-9IM	
 	
 	
 	
 	
 	
   
LXXXXX  
 4[[ 
 
199aJJqMMA     	
 	
 	
 	
  
RK:GrM:fl        	
 	
 	
 
L
Q"6a26"q&B,	
 	
 	
 	
 	
 	
 	
 	
 	
   
L
3v{3w!|	
 	
 	
 	
 	
 	
 	
 	
 	
 	
    r8   c                    | j         }t          | |||||||	|
||||||||||||           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr  r.   r   r   r   rS  r  )r   r  r   rW  rT  r,  r-  rU  r.  r/  rW  rX  rY  rZ  rT  rU  rX  rY  rZ  r[  r\  r]  r  r   s                           r6   max_pool3d_backward_shape_checkrh    s$   2 :D








+  0 ;dQh888;dQh666;dQh888;dQh7777D$(G4447D$(E2227D$(G4447D$(F33333r8   c                    | j         }t          | ||||||||	|
|ddd|||||||d           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr   Tr  r.   r   rg  )r   r  rW  rT  r,  r-  rU  r.  r/  rW  rX  rY  rX  rY  rZ  r[  r\  r]  r  r   s                       r6   re  re  <  s    * :D








			-  2 ;dQh888;dQh666;dQh888;dQh77777r8   c                 P   d } |d|          \  }}t          j        t          |          dv d            t          |          dk    r||}
}	n |d|          \  }	}
 |d|          \  }} |d|          \  }}|                     d	          }|                     d
          }|                     d          }t	          j        |           }|t           j        k    r-t          j        |                                 dk    d            nQ|t           j        k    r+t          j        |                                 dv d            nt          j        dd            t          ||||	||          }t          ||||
||          }t          | |||	|
||||||||||           |||fS )Nc                      t          j        t          |          dv  fd           |d         }t          |          dk    r|n|d         }||fS )Nr  c                      d  dS )Nzmax_pool2d: r  r<   r  s   r6   r[   zEmax_pool2d_checks_and_compute_shape.<locals>.unpack.<locals>.<lambda>~  r  r8   r   r   r  r  s   `   r6   r  z3max_pool2d_checks_and_compute_shape.<locals>.unpack{  r  r8   r  r  c                      dS )NzOmax_pool2d: stride must either be omitted, a single int, or a tuple of two intsr<   r<   r8   r6   r[   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r  r8   r   r   r  r  r$  r  r   r  c                      dS )NzMnon-empty 4D (batch mode) tensor expected for input with channels_last layoutr<   r<   r8   r6   r[   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  s    c r8   r  c                      dS )Nz9non-empty 3D or 4D (batch mode) tensor expected for inputr<   r<   r8   r6   r[   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>      O r8   Fc                      dS )NzAUnsupported memory format. Supports only ChannelsLast, Contiguousr<   r<   r8   r6   r[   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  s    W r8   )rL   r]   r   r   rF   r   r  r{   r   r(  r)  )r   r  r   r  r  r  r  r,  r-  r.  r/  r0  r1  rT  rU  r  r2  r3  r   r  r  s                        r6   r  r  r  s      VM;//FB	LFy aa   6{{aRB&))B	7++JD$!6*h77Iy**R..K**R..KBJ/66M+++IIKK1cc	
 	
 	
 	
 
%1	1	1IIKK6!OO	
 	
 	
 	

 	WW	
 	
 	

 (Rr9iXXL&z2tRIVVK



  $ k11r8   c                 P    t          |||||          \  }t          j        j         j        k     fd           |j        fd}	 |	             |	|           t          j                  }
t          j        j        j        j	        |
          S )Nc                  (    dj          d j          S )NzExpected dtype z  for `gradOutput` but got dtype r   r  s   r6   r[   z7meta_max_pool2d_with_indices_backward.<locals>.<lambda>  s    a$*aakN_aa r8   c                     t          | dz
             t          | dz
             t          | dz
             d S )Nr.   r   r   )r  )r  r8  r   r  r  s    r6   _check_dim_sizez>meta_max_pool2d_with_indices_backward.<locals>._check_dim_size  sQ    q$q,777q$q,777q$q+66666r8   rv  )
r  rL   r]   rT   r   rF   r   r   r   rx   )r  r   r  r   r  r  r  r   r  ru  r   r8  r   r  r  s   ``         @@@@r6   %meta_max_pool2d_with_indices_backwardrv    s     	,k67Hi	 		
 
L
k''aaaaa  
 L9D7 7 7 7 7 7 7 7
 OK   OG/55M;
j{#	   r8   c                    t          | |||||          \  }}}|                                 dk    r|                     d          nd}	t          j        |           }
|                                 dk    r|||g}n|	|||g}t          j        || j        | j        |
          t          j        |t
          j	        | j        |
          fS r  )
r  r{   r   rF   r   rL   r   rT   rx   r   r  s               r6   meta_max_pool2d_with_indicesrx    s     	,{FGXy	 		
  %yy{{a//UZZ^^^QF/66Myy{{a\;7\;?+<'		
 	
 	
 	+<'		
 	
 	
 r8   c           	         
 t          j         j        dv  fd            j        }t          |dz
  |          D ]G
t          j                             
          dk    d                                  d
 d           Ht          j        t                    dk    d	            t          j        t          |          dk    d
                                 d          }                     d                               d          |dk    r                     d          }nd}t          j         j        j        k    d            t          j        j        dk    fd                               d          }                    d          }                    d          
t          j        ||k    d           t          j        ||k    d            t          j        
dk    
fd           t          j        |d         d         z   dz
  k    fd           t          j        |d         d         z   dz
  k    fd                                            dk    r|||d         |d         g}	n||d         |d         g}	t          j        |	 j         j	                  t          j        |	t           j
         j	                  fS )Nr  c                      d j          S )Nz:fractional_max_pool2d: Expected 3D or 4D tensor, but got: r  r   s   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>  s    XTYXX r8   r.   r   z_fractional_max_pool2d: Expected input to have non-zero  size for non-batch dimensions, but got rt  z emptyr   c                      dS )NzNfractional_max_pool2d: kernel_size musteither be a single int or tuple of Intsr<   r<   r8   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>#  	     2 r8   c                      dS )NzOfractional_max_pool2d: output_size must either be a single int or tuple of Intsr<   r<   r8   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>(  r|  r8   r$  r  r   r  r   c                      dS )Nz6Expect _random_samples to have the same dtype as inputr<   r<   r8   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>6  s    H r8   c                      d j          S )Nz1Expect _random samples to have 3 dimensions got, r  )random_sampless   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>:  s    YNDWYY r8   z=Expect _random_samples.size(0) no less then input batch size.c                      dS )Nz<Expect _random_samples.size(1) equals to input channel size.r<   r<   r8   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>F      N r8   c                      d  dS )Nz/Expect _random_samples.size(2) equals to 2 got .r<   )r   s   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>H  s    !WST!W!W!W r8   c                       dd          d  S )Nz%fractional_max_pool2d: kernel height r   z' is too large relative to input height r<   )input_heightr  s   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>L  s    }A}}o{}} r8   c                       dd          d  S )Nz$fractional_max_pool2d: kernel width r   z& is too large relative to input width r<   )input_widthr  s   r6   r[   z,meta_fractional_max_pool2d.<locals>.<lambda>P  s    z{1~zzmxzz r8   r  )rL   r]   r   r   r   r   rT   r{   r   rx   r   )r   r  rm  r  r   input_channelsinput_batchr  cr   r   r  r  s   `` `      @@@r6   meta_fractional_max_pool2dr    sb   	L	VXXXX   9D4!8T"" 
 
IIaLL1^7;yy{{^ ^TU^ ^ ^	
 	
 	
 	
 
LKA	2 	2  
 
LKA	2 	2   YYr]]N99R==L))B--Kqyyiill	L
n**HH   
Lq YYYY  
 	AAAAAA	L	[G   
L	^NN   
LaWWWWXXX	LAQ'!+|;}}}}}   
LAQ'!+{:zzzzz  
 xxzzQ^[^[^LAA? 	*;	
 	
 	

 	+;	
 	
 	
 r8   c                 T   t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        | pt          |          dv d            |s|n|d         }	|s|nt          |          dk    r|	n|d         }
|s|nt          |          dk    r|	n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        | j        d	v d
            | j        dk    r|                     d          nd}|                     d          }|                     d          }|                     d          }|                     d          }t          ||||	||          }t          ||||
||          }t          ||||||          }t          | |||||	|
|||||||||||||d           | j        dk    o!t          j        |           t           j	        k    }| j        dk    rQ| 
                    d          }|                                 o|                    t           j	                  }||||f}n|||||f}|                     |          }|                     |t           j                  }|r@|                    t           j	                  }|                    t           j	                  }||fS )NrC  c                      dS NzMmax_pool3d: kernel_size must either be a single int, or a tuple of three intsr<   r<   r8   r6   r[   z.meta_max_pool3d_with_indices.<locals>.<lambda>r      _ r8   r   r   r   c                      dS NzQmax_pool3d: stride must either be omitted, a single int, or a tuple of three intsr<   r<   r8   r6   r[   z.meta_max_pool3d_with_indices.<locals>.<lambda>z      c r8   c                      dS NzImax_pool3d: padding must either be a single int, or a tuple of three intsr<   r<   r8   r6   r[   z.meta_max_pool3d_with_indices.<locals>.<lambda>      [ r8   c                      dS NzJmax_pool3d: dilation must be either a single int, or a tuple of three intsr<   r<   r8   r6   r[   z.meta_max_pool3d_with_indices.<locals>.<lambda>  rn  r8   rM  c                      dS rO  r<   r<   r8   r6   r[   z.meta_max_pool3d_with_indices.<locals>.<lambda>  rP  r8   r!  r  r$  r  r   zmax_pool3d_with_indices()r  r   r   )rL   r]   r   r   r   r(  rS  rF   r   r  r  rZ  r   r   r"  )r   r  r   r  r  r  rT  r,  r-  rU  r.  r/  rW  rX  rY  rZ  rT  rU  r  rW  rX  rY  rZ  r[  r\  r]  r  input_channels_last_checkr   r   r   s                                  r6   meta_max_pool3d_with_indicesr  f  s    
LKF"__   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+cc   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LG[[   
B7||q  gajB7||q  gajB	LH\\   I ]]a//		Xa[I ]]a//		Xa[I	L
fKK  
  %zQUZZ^^^AFjjnnGJJrNNEjjnnGZZ^^F BIyIIE"7BB	9MMG!&"b"iKKF








#+  2 	
aXE7>>%BXX  zQ$)OOA$6$6!)77999
'550 6 
 
 	
 eWf5		WeWf=	
//)
$
$Cooiu{o;;G Cff5#9f::**5+A*BB<r8   c                    t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }	t          |          dk    r|n|d         }
t          j        | pt          |          dv d            |s|n|d         }|s|	nt          |          dk    r|n|d         }|s|
nt          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        |j        d	v d
            |                    d          }|                    d          }|                    d          }|                    d          }|                     d          }|                     d          }|                     d          }t          || ||||	|
|||||||||||||||d           |j        dk    o!t          j        |          t           j        k    }|j        dk    rJ|	                    d          }|
                                 o|
                    t           j                  }|                    |j                  }|r |                    t           j                  }|S )NrC  c                      dS r  r<   r<   r8   r6   r[   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r8   r   r   r   c                      dS r  r<   r<   r8   r6   r[   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r8   c                      dS r  r<   r<   r8   r6   r[   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r8   c                      dS r  r<   r<   r8   r6   r[   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  rn  r8   rM  c                      dS rO  r<   r<   r8   r6   r[   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  rP  r8   r  r$  r  r   z"max_pool3d_with_indices_backward()r!  r  r   )rL   r]   r   r   r   rh  rF   r   r  r  rZ  r   r   r"  )r  r   r  r   r  r  r  r   rT  r,  r-  rU  r.  r/  rW  rX  rY  rZ  rT  rU  rW  rX  rY  rZ  r[  r\  r]  r  r  r  s                                 r6   %meta_max_pool3d_with_indices_backwardr    s    
LKF"__   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+cc   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LG[[   
B7||q  gajB7||q  gajB	LH\\   I ]]a//		Xa[I ]]a//		Xa[I	L
fKK  
 jjnnGJJrNNEjjnnGZZ^^FR  Er""Gb!!F#








,/  6 	
aXE7>>%BXX  zQ$)OOA$6$6!)77999
'550 6 
 
 	 --J I]]1G]HH
r8   gridc                 "    t          j         j        j        k     fd           t          j         j        t           j        k    oj        t           j        k     fd           t          j         j        d         j        d         k     fd           t          j        j        d          j        dz
  k     fd           t          d j                  D ]*t          j         j                 dk     fd           +d S )	Nc                  (    dj          d j          S )NzNgrid_sampler(): expected input and grid to be on same device, but input is on z and grid is on r  r  r   s   r6   r[   z+check_grid_sampler_common.<locals>.<lambda>9  s0    A\A A37;A A r8   c                  (    dj          d j          S )NzTgrid_sampler(): expected input and grid to have torch.strided layout, but input has z and grid has )rw   r  s   r6   r[   z+check_grid_sampler_common.<locals>.<lambda>@  s0    CC C59[C C r8   r   c                  (    dj          d j          S )NzZgrid_sampler(): expected grid and input to have same batch size, but got input with sizes  and grid with sizes r2  r  s   r6   r[   z+check_grid_sampler_common.<locals>.<lambda>G  s0    O %O OBF*O O r8   r   r   c                  .    dj         dz
   d j         S )Nz+grid_sampler(): expected grid to have size r   z, in last dimension, but got grid with sizes )r   r   r  s   r6   r[   z+check_grid_sampler_common.<locals>.<lambda>N  s.    ?%*q. ? ?26*? ? r8   c                       dj          d  dS )NzYgrid_sampler(): expected input to have non-empty spatial dimensions, but input has sizes rt  ru  r2  r  s   r6   r[   z+check_grid_sampler_common.<locals>.<lambda>W  r  r8   )rL   r]   rx   rw   r  r   r   r   )r   r  r   s   ``@r6   check_grid_sampler_commonr  6  st   	L#	
 	
 	
 	
 	
   
L%F$+*F	
 	
 	
 	
 	
   
LA$*Q-'	
 	
 	
 	
 	
   
L
2%*q.(	
 	
 	
 	
 	
   1ej!! 
 
KNQ    	
 	
 	
 	

 
r8   c                       e Zd ZdZdZdZdS )GridSamplerInterpolationr   r   r   N)rp   
__module____qualname__BILINEARNEARESTBICUBICr<   r8   r6   r  r  ^  s        HGGGGr8   r  interpolation_modec                      t          j         j        dk    o j        j        k     fd           t          j         j        dk    o|t          j        j        k     d            d S )Nr!  c                  (    dj          d j          S )Nzdgrid_sampler(): expected 5D input and grid with same number of dimensions, but got input with sizes r  r2  r  s   r6   r[   z'check_grid_sampler_3d.<locals>.<lambda>g  s+    149K1 1$(J1 1 r8   c                      dS )Nz<grid_sampler(): bicubic interpolation only supports 4D inputr<   r<   r8   r6   r[   z'check_grid_sampler_3d.<locals>.<lambda>r  r  r8   )rL   r]   r   r  r  r  )r   r  r  s   `` r6   check_grid_sampler_3dr  d  s    	L
a3EJ$)3	
 	
 	
 	
 	
   
LJ!O M"&>&F&LL	
 	ON    r8   c                     |d         }|r!t          j        |t           j                  }nd }t          j        |t           j                  }	||	fS Nr   r   )rL   r  r   r   
r  r   r  r  padding_modealign_cornersr  input_requires_gradr  	grad_grids
             r6   grid_sampler_2d_backward_metar  v  sX     &a. %e5;RSSS


 U5LMMMI	""r8   c                     t          | |           t          | ||           | j        d         }| j        d         }|j        d         }|j        d         }|j        d         }	|                     |||||	f          S )Nr   r   r   r.   )r  r  r   r   )
r   r  r  r  r  r  Cout_Dout_Hout_Ws
             r6   grid_sampler_3dr    s{     eT***%'9:::AAAAJqMEJqMEJqME??Aq%6777r8   r  c                     t          ||           t          |||           |d         }|r!t          j        |t          j                  }nd }t          j        |t          j                  }	||	fS r  )r  r  rL   r  r  r   r  s
             r6   grid_sampler_3d_backwardr    s     eT***%'9:::%a. %!?
 
 


 
 U5STTTIy  r8   c                     |                     dd           }|st          j        |          }||d<   t          j        | g|R i |S )NrT   )rS   rF   	get_dtyperL   r   )r   r  rH   r  rT   s        r6   fullr    sU    JJw%%E ,
++F7O;t-d---f---r8   c                 n   |t           j        k    rt          j        |d u d            t          j        d|| j        n|||| j        n||          }| j        rN|                    |                                 | 	                                | 
                                           n;|                    |                                 |                                 d           |                    d           |S t          j                            | |||||          }|                    d           |S )Nc                      dS )Nz9memory format option is only supported by strided tensorsr<   r<   r8   r6   r[   zzeros_like.<locals>.<lambda>  rp  r8   r   r  Tr  )rL   
sparse_coor]   r   rT   rx   	is_sparsesparse_resize_and_clear_r   
sparse_dim	dense_dimr{   _coalesced_r+   r   r$  fill_)r   rT   rw   rx   ry   r   r  s          r6   r  r    sA    !!!T!OO	
 	
 	

 k %$**5"(.4;;f!
 
 
 > 	E((		T__..0@0@    ((dhhjj!DDD

/
!
!# "  C IIaLLLJr8   rv   c                    |t          j                    }|t          j                    }|t           j        }t          j        | ||||          S r  rL   r}   get_default_devicer  r   r   rT   rw   rx   ry   rz   s         r6   	meta_onesr    Z     }'))~)++~;E&J   r8   c                    |t          j                    }|t          j                    }|t           j        }t          j        | ||||          S r  r  r  s         r6   
meta_zerosr    r  r8   c                 *    t          j        |           S r3   rF   clone_preserve_strides)r   r%  r{   r   s       r6   meta_select_scatterr        '---r8   c                 *    t          j        |           S r3   r  )r   r%  r{   rr   rq   steps         r6   meta_slice_scatterr    r  r8   dim_post_exprwrap_scalarc                     |dk    r|sJ d}| }|dz
  }| |k     s| |k    rJ d|  d| d| d            | dk     r| |z  } | S )Nr   r   zdim z out of bounds (rm   rn   r<   )r{   r  r  r   r  s        r6   r   r   "  s~    {.C
!
Cc		S3YYY)R)R)RS)R)RC)R)R)RYY'
Qww}Jr8   c                 P    |                                  dk    rdn| j        |         S r  r  )r  r{   s     r6   ensure_nonempty_sizer  .  s"    111!'#,.r8   c                 p    t                                           d          }t                                          d          }t          j        ||k    d            t	          |          D ]Ck    r;t          j        t                    t                     k     fd           Dd S )Nr   c                      dS )NzDIndex tensor must have the same number of dimensions as input tensorr<   r<   r8   r6   r[   z$gather_shape_check.<locals>.<lambda>8  s    V r8   c                  :    d dj          dj          d  z   S )Nz!Size does not match at dimension z expected index  to be no larger than self  apart from dimension r2  )r{   r   r   r   s   r6   r[   z$gather_shape_check.<locals>.<lambda>>  s6    \A\\u{\\W
WWRUWWX r8   )r  r{   rL   r]   r   r  )r   r{   r   	self_dims
index_dimsr   s   ```  @r6   gather_shape_checkr  3  s    DHHJJ""IUYY[[!$$J	LZVV   9  88L$UA..2FtQ2O2OOX X X X X X X   r8   c                 p   ddl m} t          ||                                           } |                                dk              }|sQt          j        j        t
          j        k    pj        t
          j	        k    fd           t          | |           |                     j                  S )Nr   r  c                      d j          S )Nz8gather(): Expected dtype int32/int64 for index, but got r   r   s   r6   r[   zmeta_gather.<locals>.<lambda>L  s    \u{\\ r8   )r  r  r   r{   r   rL   r]   rT   r   r   r  r   r   )r   r{   r   sparse_gradr  wrapped_dimis_index_emptys     `    r6   meta_gatherr  C  s    DDDDDD dhhjj11K#^EKKMMQ$677N 5K5:%A	)A\\\\	
 	
 	
 	4e444>>%+&&&r8   c                     |r@| dk    rdS | dk    rdS | dk    rdS | dk    rdS | d	k    rd
S t          j        dd            d S | dk    rdS | dk    rdS t          j        dd            d S )Nr  
REDUCE_ADDr  REDUCE_MULTIPLYmeanREDUCE_MEANamaxREDUCE_MAXIMUMaminREDUCE_MINIMUMFc                      dS )Nz=reduce argument must be either sum, prod, mean, amax or amin.r<   r<   r8   r6   r[   z#get_operator_enum.<locals>.<lambda>a  s    S r8   addmultiplyc                      dS )Nz/reduce argument must be either add or multiply.r<   r<   r8   r6   r[   z#get_operator_enum.<locals>.<lambda>i  s    $U r8   r[  )reduce_use_new_optionss     r6   get_operator_enumr  S  s     e<$$ =####SS	
 	
 	
 	e<
""$$UUUVVVr8   c                 *    ddl m}  ||                                dk              r@t          j        |j        t          j        k    p|j        t          j        k     fd           |(t          j        |j        |j        k     fd           d S d S )Nr   )rE  c                        dS )Nz((): Expected dtype int32/int64 for indexr<   method_names   r6   r[   z,scatter_gather_dtype_check.<locals>.<lambda>t  s    {LLL r8   c                        dS )Nz0(): Expected self.dtype to be equal to src.dtyper<   r  s   r6   r[   z,scatter_gather_dtype_check.<locals>.<lambda>z  s    {TTT r8   )r  rE  r   rL   r]   rT   r   r   )r  r   r   src_optrE  s   `    r6   scatter_gather_dtype_checkr  n  s    CCCCCC}U[[]]a'(( 
K5:%A	)ALLLL	
 	
 	

 J'-'TTTT	
 	
 	
 	
 	
 r8   c                 "    t          | d          S r*  )r  r   s    r6   ensure_nonempty_dimr  ~  s    sA;;r8   c                 z    ddl m}  |                                dk              rd S t          j        t                                                     t                                                    k    d            d}t                                                     }t          |          D ]1}t          |          }|k    r|t           |          k    rd} n2|s<:t          |          D ]*}t          |          }|t          |          k    rd} n+vt          j        t                                                     t                                                    k    d            t          j        |  fd           d S t          j        |  fd           d S )	Nr   r  c                      dS NzCIndex tensor must have the same number of dimensions as self tensorr<   r<   r8   r6   r[   z%scatter_shape_check.<locals>.<lambda>  r  r8   FTc                      dS r  r<   r<   r8   r6   r[   z%scatter_shape_check.<locals>.<lambda>  s    Y r8   c                  D    dj          dj          d  dj          z   S )NExpected index r  r  z and to be no larger than src r2  )r{   r   r   r  s   r6   r[   z%scatter_shape_check.<locals>.<lambda>  s7    ZekZZdjZZYsYY'-YYZ r8   c                  4    dj          dj          d  z   S )Nr  r  r  r2  )r{   r   r   s   r6   r[   z%scatter_shape_check.<locals>.<lambda>  s,    ZekZZdjZZ,s,,- r8   )	r  r  r   rL   r]   r  r{   r   r  )	r   r{   r   r  r  is_wrong_shaper  r   index_d_sizes	   ````     r6   scatter_shape_checkr    s-   DDDDDD~ekkmmq()) 	LDHHJJ''+>uyy{{+K+KKUU  
 N#DHHJJ//I 9  +E15588.tQ7777!NE 8
  g1y!! 	 	A/q99L27A>>>>!% ? 

++/B599;;/O/OOYY	
 	
 	
 	Z Z Z Z Z Z Z	
 	
 	
 	
 	
 	- - - - - -	
 	
 	
 	
 	
r8   c                     t          ||                                           }t          d| ||           t          | |||           |t	          ||           d S d S )Nscatter)r   r{   r  r  r  )r   r{   r   r%  r  r  r  s          r6   scatter_meta_implr    se     dhhjj11Ky$s;;;k5#666'?33333 r8   c                 \    t          | |||d           |                     | j                  S Nr  r  r   r   r   r{   r   r%  s       r6   meta_scatter_addr#    s,    dCU333>>$*%%%r8   c                 ,    t          | |||d           | S r   r  r"  s       r6   meta_scatter_add_r&    s    dCU333Kr8   c                     t          |t          j                  r|nd }t          | ||||           |                     | j                  S r3   )rf   rL   r   r  r   r   r   r{   r   src_or_valuerc  r%  s         r6   meta_scatterr*    sH     %\5<@@
J,,dCdCV444>>$*%%%r8   c                 h    t          |t          j                  r|nd }t          | ||||           | S r3   )rf   rL   r   r  r(  s         r6   meta_scatter_r,    s9     %\5<@@
J,,dCdCV444Kr8           queryr   r  	dropout_p	is_causalreturn_debug_maskrs  c           	         |                      d          }|                      d          }|                      d          }	|                      d          }
|                     d          }|                     dd          }t          j        |                              dd          }t          j        |||	ft          j        | j                  }|rX|
dk    rdnd}t          j        |	|z            }|dk    rd}n|dk    rd}t          j        |||	|f| j	        | j                  }n!t          j        d| j	        | j                  }t          j
        j        rat          j                                        rCt          j        d	t          j        d
          }t          j        d	t          j        d
          }nBt          j        dt          j        d
          }t          j        d	t          j        d
          }||d d |	||||f	S )Nr   r   r   r.   r  @         r<   ru   )r   r  rL   r   r   rP   rx   r  ceilrT   versionhipr   rl  r   r'  )r.  r   r  r/  r0  r1  rs  r   	num_headsmax_seqlen_batch_qhead_dimmax_seqlen_batch_kquery_t	attention	logsumexpblocksize_cmax_seqlen_k
debug_maskseedoffsets                       r6   (meta__scaled_dot_product_flash_attentionrE    s    AJ

1IAzz!}}H!ooa##G ))33Aq99I	Y 23k|  I  L%]]ccy!3k!ABB$$LL3&&L[$6E+<
 
 


 [%+elKKK
 } DUZ4466 D{2UZ???Ruz&AAA{Ael6BBBRu|FCCC 	
 
r8   	res_shape.c                     t           j                  k    r?                     dd          }t          j        |                              dd          }nt          g d fdd          fdD             }fdt          t                              D             }t          j        | j	         j
        	                              |          }|S )
Nr   r   )r   r   r   r.   c                 8                                     |          S r3   r  )idxr.  s    r6   r[   z,alloc_with_matching_layout.<locals>.<lambda>0  s    %,,..*= r8   Tr   c                      g | ]
}|         S r<   r<   )rB   rI  rF  s     r6   rE   z.alloc_with_matching_layout.<locals>.<listcomp>2  s    >>>S)C.>>>r8   c                 :    g | ]}                     |          S r<   r   )rB   r   	dim_orders     r6   rE   z.alloc_with_matching_layout.<locals>.<listcomp>3  s%    KKK++KKKr8   r  )r\   r   r  rL   r   sortedr   r   r   rT   rx   r   )r.  rF  r=  r  permuted_shapefinal_permuterL  s   ``    @r6   alloc_with_matching_layoutrP  '  s     U[Y&&//!Q''w''11!Q77LL====t
 
 
	 ?>>>I>>>KKKKU3y>>5J5JKKKk%+el
 
 

'-
 
  	 Jr8   	attn_biascompute_log_sumexpc	           	         |                      d          }	|                      d          }
|                      d          }|                     d          }|                     d          }|	|
||f}t          | |          }t          j        |	|
|dft          j        | j                  }t          j        dt          j        d          }t          j        dt          j        d          }||d d ||||d f	S Nr   r   r   r   r  r<   ru   r   rP  rL   r   rP   rx   r   )r.  r   r  rQ  rR  r/  r0  r1  rs  r  r  S_QS_KVD_VrF  r  
logsum_exprC  rD  s                      r6   (meta__scaled_dot_product_cudnn_attentionrZ  ;  s     	

1A

1A
**Q--C88A;;D
**R..CAsC I
$UI
6
6C	
AsAk|  J ;rF;;;D[5:f===F 	
 
r8   c           	         |                      d          }|                      d          }	|                      d          }
|                     d          }|                     d          }||	|
|f}t          | |          }t          j        ||	|
ft          j        | j                  }t          j        dt          j        d          }t          j        dt          j        d          }||d d |
|||d f	S rT  rU  )r.  r   r  rQ  r/  r0  r1  rs  r  H_QrV  rW  rX  rF  r  rY  rC  rD  s                     r6   5meta__scaled_dot_product_fused_attention_overrideabler]  g  s     	

1A
**Q--C
**Q--C88A;;D
**R..CCc"I
$UI
6
6C	
Ck|  J ;rF;;;D[5:f===F 	
 
r8   rv  r?  	cum_seq_q	cum_seq_kmax_qmax_kphilox_seedphilox_offsetc                 t   t          j        |                    dd                                        dd          }t          j        |                    dd                                        dd          }t          j        |                    dd                                        dd          }|||fS r  )rL   r   r  )rv  r.  r   r  r   r?  r^  r_  r`  ra  r/  r0  rb  rc  rs  grad_qgrad_kgrad_vs                     r6   'meta__scaled_dot_product_flash_backwardrh    s    , eooa3344>>q!DDFcmmAq1122<<QBBFeooa3344>>q!DDF66!!r8   	attn_maskc                 *   |                      d          }|                      d          }|                      d          }	t          j        |           }
t          j        ||	|ft          j        | j                                      dd          }|
|fS )Nr   r   r   r  )r   rL   r   r   rP   rx   r  )r.  r   r  r/  r0  ri  rs  r   r9  r:  r>  r?  s               r6   0meta__scaled_dot_product_flash_attention_for_cpurk    s     AJ

1IA ''I	

 k|   i1oo  	 r8   c
                 D   t          j        |                                d|j        |j                  }
t          j        |                                d|j        |j                  }t          j        |                                d|j        |j                  }|
||fS )Nr   r   r   r.   r  )rL   empty_permutedr   rT   rx   )rv  r.  r   r  r   r?  r/  r0  ri  rs  re  rf  rg  s                r6   9meta__scaled_dot_product_flash_attention_for_cpu_backwardro    s    & !

k|	  F !

iz	  F !

k|	  F 66!!r8   dropout_maskc                     d } |           \   ||          \  }	}
 ||          \  }}
j         \  |	j         \  }
}}
 fd}fd}dk    s|k     rdk    r
 |            S  |            S )Nc                    |                                  dk    r|                     d          dfS |                                  dk    rd}t          |                                  dz
            D ]}|| j        |         z  }|                     ||                     d          |                     d          |                     d                    dfS | d	fS )
Nr.   r   Tr  r   r$  r  r   F)r{   r  r   r   viewr   )rC   r   r   s      r6   	ensure_4dzBmeta__scaled_dot_product_attention_math_for_mps.<locals>.ensure_4d  s    5577a<<;;q>>4''UUWWq[[J15577Q;'' ) )agaj(

66*affRjj!&&**affRjjII4OOe8Or8   c                                          j                  } 	r|                               }                      f          }	ro                                dk    r|                    d          }nAt          j        d d                   |j        dd         z   }|                    |          }| |fS )Nr.   r   r$  r   r  )r   r   view_asr{   squeezer   rs  )
r   attnr   r   max_seq_lengthnum_headq_q_sizer.  
unsqueezeds
      r6   sdpa_vector_fast_mpszMmeta__scaled_dot_product_attention_math_for_mps.<locals>.sdpa_vector_fast_mps  s    ll28$$ 	%++e$$C||Z6>JKK 	(yy{{a||AU["-..AaC@yy''Dyr8   c                  x    d}                      j                  }                     | f          }||fS )Nr  r  )blocksr   r'  r   	head_sizerz  r{  r|  s      r6   sdpa_vector_2pass_mpszNmeta__scaled_dot_product_attention_math_for_mps.<locals>.sdpa_vector_2pass_mps$  sB    ll28$$||Z669$UVVL  r8   i   i   r2  )r.  r   r  ri  r/  r0  rp  rs  rt  k_rI   v_k_sizer~  r  r   r  ry  rz  r{  r|  r}  s   `              @@@@@@@r6   /meta__scaled_dot_product_attention_math_for_mpsr    s   	 	 	 Yu%%NB
IcNNEBIeEB.0h+J&)#%8 Av~q          ! ! ! ! ! ! ! ! ! 	$FVOO$8N8N$$&&&##%%%r8   c                 .   |                      dd          } |                     dd          }|                     dd          }|                     d          }|                     d          }	|                     d          }
|                    d          }t          j        ||	|
|| j        | j                  }t          j        j        r&t          j        	                                r	 |r|	nd}n|rt          j        |	dz            dz  nd}t          j        ||
|ft          j        | j                  }|                     dd          }t          j        dt          j        d	          }t          j        dt          j        d	          }||||fS )
Nr   r   r   r  r   r  r  r<   ru   )r  r   rL   r   rT   rx   r7  r8  r   rl  r  r6  rP   r   )r.  r   r  rQ  rR  r/  r0  rs  r  r  r9  Kvr  logsumexp_dimrY  rC  rD  s                    r6   ,meta__scaled_dot_product_efficient_attentionr  0  sv    OOAq!!E
--1

COOAq!!E

1A

1A

2I	BB
+aIrU\
R
R
RC} LUZ4466 L	 06Q2DK	!b&))B..!	
I}%k|  J --1

C ;rF;;;D[5:f===F
D&((r8   grad_input_maskc                    |                     d          }|                     d          }|                     d          }|                     d          }|                     d          }|                     d          }t          j        ||||fd|j        |j                  }t          j        ||||fd|j        |j                  }t          j        ||||fd|j        |j                  }d }||
d         r~|                     d          }|dz  dk    r|n
|dz   |dz  z
  }t          |                                           }||d<   t          j        ||j        |j                  }|d	d |f         }||||fS )
Nr   r   r   r.   rm  r  r   rU  .)r   rL   rn  rT   rx   r   r   )rv  r.  r   r  rQ  r   r?  rb  rc  r/  r  r0  rs  r   r9  r`  r;  
head_dim_vra  re  rf  rg  	grad_biaslastDimlastDimAligned	new_sizess                             r6   +meta__scaled_dot_product_efficient_backwardr  ]  s   ( AJ

1IJJqMMEzz!}}HAJHHQKKE!	Yx0k|	  F !	Yx0iz	  F !	Yz2k|	  F I!3..$$$+bLA$5$57R<'TV,;V))**	&	"KY_Y5E
 
 
	 c8G8m,	669,,r8   c                     t          j        |          }t          j        |          }t          j        |          }|||fS r3   ri  )rv  r.  r   r  r   r?  rb  rc  rQ  r^  r_  r`  ra  r/  r0  rs  re  rf  rg  s                      r6   'meta__scaled_dot_product_cudnn_backwardr    sA    . e$$Fc""Fe$$F66!!r8   window_size_leftwindow_size_right	seqused_kalibi_slopesc                    ||                      d          n|                                dz
  }||                      d          n|}||                     d          n|}|                      d          }|                      d          }t          j        |           }|*t          j        |||ft          j        | j                  }n=|                      d          }t          j        ||ft          j        | j                  }|	rX|dk    rdnd}t          j        ||z            }|dk    rd}n|dk    rd}t          j        ||||f| j	        | j                  }n!t          j        d| j	        | j                  }d	\  }}t          j
        j        rat          j                                        rCt          j        d
t          j        d          }t          j        d
t          j        d          }nBt          j        dt          j        d          }t          j        d
t          j        d          }|||||fS )Nr   r   r  r   r  r3  r4  r5  NNr<   ru   r   )r   r   rL   r   r   rP   rx   r  r6  rT   r7  r8  r   rl  r   r'  )r.  r   r  r^  r_  r`  ra  r/  r0  r1  rs  r  r  r  r  r   r:  r<  r9  r;  r>  r?  total_qr@  rA  rB  rC  rD  s                               r6   meta__flash_attention_forwardr    s?   4 #,"3A9J9JQ9NJ*3*;A(1(9!u

2Izz"~~H  ''IK$67+<
 
 
		 **Q--K EL
 
 
	  L%]]ccy!3k!ABB$$LL3&&L[$6E+<
 
 


 [%+elKKK
 LD&} DUZ4466 D{2UZ???Ruz&AAA{Ael6BBBRu|FCCC r8   c                     t          j        |          }t          j        |          }t          j        |          }|||fS r3   ri  )rv  r.  r   r  r   r?  r^  r_  r`  ra  r/  r0  rb  rc  rs  r  r  
grad_querygrad_key
grad_values                       r6   meta__flash_attention_backwardr    sA    0 !%((J$$H!%((Jx++r8   cu_seqlens_qcu_seqlens_kmax_seqlen_qrA  custom_mask_typecausal_diagonalseqlen_kwindow_sizec                    |                      d          }|                      d          }|                     d          }|                      d          }|                     d          }t          j        ||||| j        | j                  }||                     d          dz
  n|}|}||J |}||n|}|
rt          j        |dz            dz  nd}t          j        |||ft          j        | j                  }t          j        dt          j        d          }t          j        dt          j        d          }||||||fS )	Nr   r   r  r   r  r  r<   ru   )	r   rL   r   rT   rx   r  r6  rP   r   )r.  r   r  r7  r  r  r  rA  r/  r  rR  rs  r  r  r  r  r  r  r9  r  r  logsumexp_batch_dimactual_max_seqlen_qactual_max_seqlen_kr  rY  rC  rD  s                               r6   !meta__efficient_attention_forwardr  %  sZ   , 	

1A

1AA

2I	BB
+aIrU\
R
R
RC7C7O,++A..22VW'''**6*B,,4FM	%*++b00A  	i7k|  J ;rF;;;D[5:f===F
D&*=?RRRr8   bias_requires_gradnum_splits_keyshared_storage_dqdkdvc                    |rt          j        |j        d         |j        d         k    d            t          j        |j        d         |j        d         k    d            t          j        g |j        dd         d|j        d         |j        d         R |j        |j                  }|                    d	d          }|                    d	d          }|                    d	d
          }n<t          j        |          }t          j        |          }t          j        |          }||                    d          }|dz  dk    r|n
|dz   |dz  z
  }t          |                                          }||d<   t          j        ||j        |j                  }|dd |f         }nt          j        d|j                  }||||fS )Nr   c                      dS )Nz,seqlen must match for `shared_storage_dqdkdvr<   r<   r8   r6   r[   z4meta__efficient_attention_backward.<locals>.<lambda>u  s    B r8   r.   c                      dS )Nz3embedding dim must match for `shared_storage_dqdkdvr<   r<   r8   r6   r[   z4meta__efficient_attention_backward.<locals>.<lambda>y  s    I r8   r   r  r   r  r$  r   rU  .r<   r  )
rL   r]   r   r   rT   rx   r  r   r   r   )rv  r.  r   r  r7  r  r  r  rA  r?  r/  rb  rc  r  r  rs  r  r  chunkr  r  r  r  r  r  r  s                             r6   "meta__efficient_attention_backwardr  Y  s   2  -KNcil*BB	
 	
 	
 	KNcil*II	
 	
 	
 Eek!B$EEEKOEU[_EE+<
 
 

 \\"a((
<<A&&\\"a((

%e,,
#C((%e,,
))B--$+bLA$5$57R<'TV,;V%%	&	"K	DKPPP	c8G8m,		K5<888	xY66r8   scale_ascale_bscale_resultuse_fast_accumc                 4
    d }t          j                                         dk    o                                dk     fd           t          j         | j                  o |j                   fd           t	                     dk    r.d }	d }
d }t          j         |	                                           p
 |            fd	           t          j         |
                                          p
 |          fd
           t          j                             d          dz  dk     fd           t          j                            d          dz  dk    o                    d          dz  dk    fd            j        \  }                    d          j        t           j        k    oj        t           j        k    p)j        t           j	        k    oj        t           j	        k    }
                                dk    rX
                                dk    r@t          j        j        t           j        k    oj        t           j        k    d            nF|rj        t           j	        k    rd}|dz  }nd}d}d } |||          } ||d          dz  }| ||          z  |z  | ||          z  |z  
                                k    rj
                                k    rRt          j                                        d            t          j                                        d            n^t          j        dfd           nAt          j        j        t           j        k    oj        t           j        k    d            t          j                                        dk    o                                dk    fd                               d          k    r                    d          dk    ro                    d          dk    rV                    d          k    r=t          j                                        o                                d            nt          j        dfd           ||n j        }t          j                             d                              d          | j                  S )Nc                 x    | t           j        t           j        t           j        t           j        t           j        fv S r3   )rL   rY  float8_e5m2float8_e4m3fnuzfloat8_e5m2fnuzfloat4_e2m1fn_x2r   s    r6   is_fp8_or_fp4_typez*meta_scaled_mm.<locals>.is_fp8_or_fp4_type  s2    !!"
 
 	
r8   r   c                  \    d                                  d                                   S )Nz%Inputs must be 2D but got self.dim()=z and mat2.dim()=r   rD  r   s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s-    `

``TXT\T\T^T^`` r8   c                  (    dj          d j          S )Nz?Expected both inputs to be fp8 or fp4 types but got self.dtype=z and mat2.dtype=r   r  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s     zRVR\zznrnxzz r8   r   c                 >    | d         | d         k    o| d         dk    S r  r<   r  s    r6   is_row_majorz$meta_scaled_mm.<locals>.is_row_major  s"    !9vay(;VAY!^;r8   c                 2    | d         dk    o| d         dk    S r  r<   r  s    r6   is_col_majorz$meta_scaled_mm.<locals>.is_col_major  s    !9>3fQi!m3r8   c                 f    |                      d          dk    p|                      d          dk    S r  r   )	tensor_2ds    r6   has_zero_dimz$meta_scaled_mm.<locals>.has_zero_dim  s/    >>!$$)CY^^A->->!-CCr8   c                  2    d                                   S )Nz#self must be row_major, got stride r  r   s   r6   r[   z meta_scaled_mm.<locals>.<lambda>      I$++--II r8   c                  2    d                                   S )Nz#mat2 must be col_major, got stride r  rD  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  r  r8   r   rU  r   c                  4    d                      d           S )NzBExpected self.size(1) to be divisible by 16, but got self.size(1)=r   r   r   s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s     gY]YbYbcdYeYegg r8   c                      d j          S )Nz?Expected both dimensions of mat2 to be divisible by 16 but got r2  r  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s    bVZV`bb r8   c                      dS )NzNFor tensorwise scaling, both scale_a and scale_b must be float (fp32) tensors.r<   r<   r8   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s    h r8   r  r4  c                     | |z   dz
  |z  S r*  r<   r  s     r6   ceil_divz meta_scaled_mm.<locals>.ceil_div  s    A	a''r8   r  c                      dS )Nzscale_a must be contiguousr<   r<   r8   r6   r[   z meta_scaled_mm.<locals>.<lambda>      8 r8   c                      dS )Nzscale_b must be contiguousr<   r<   r8   r6   r[   z meta_scaled_mm.<locals>.<lambda>  r  r8   Fc            	      j    d  d                                  d d                                  d	S )NzTInvalid blockwise scaling configuration. For blockwise scaling, scale_a should have  elements, got z, scale_b should have r  r  )expected_a_sizeexpected_b_sizer  r  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  sd    bFUb bfmfsfsfufub b/>b bOV}}b b b r8   c                      dS )NzKFor rowwise scaling, both scale_a and scale_b must be float (fp32) tensors.r<   r<   r8   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s    e r8   c                  \    d                                  d                                 S )NzLFor non-tensorwise scaling, scale tensors must be 2D, but got scale_a.dim()=z and scale_b.dim()=r   r  r  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>  s2    ~Y`YdYdYfYf~~nunynyn{n{~~ r8   c                      dS )Nz@Both scale_a and scale_b must be contiguous for rowwise scaling.r<   r<   r8   r6   r[   z meta_scaled_mm.<locals>.<lambda>%  s    ^ r8   c                      d  d d                     d           d                     d           d                     d           d                     d           dS )	Nz}Invalid scaling configuration. For tensorwise scaling, both scales should be scalar. For rowwise scaling, scale_a should be (z, 1), scale_b should be (1, z). Got scale_a.size()=(r   rm   r   z) and scale_b.size()=(rn   r   )r^  r  r  r  s   r6   r[   z meta_scaled_mm.<locals>.<lambda>+  s    UCDU UbcU U 07||AU U CJ,,q//U U 07||A	U U CJ,,q//	U U U r8   r  )rL   r]   r{   rT   r   r   r   r   float8_e8m0fnurY  r   rV  rZ  r   rx   )r   rD  r  r  r7  r  r9  r  r  r  r  r  _kis_blockwise_scalingblock_size_kblock_size_mnr  num_k_blockspadded_num_k_blocks
_out_dtyper  r  r^  r  s   ````                @@@@r6   meta_scaled_mmr    s   
 
 
 
L

a+DHHJJ!O`````   
L4:&&I+=+=dj+I+Izzzzz  
 4F""	< 	< 	<	4 	4 	4	D 	D 	D 	L''=<<+=+=IIII	
 	
 	
 	L''=<<+=+=IIII	
 	
 	
 	IIaLL2"gggg	
 	
 	
 	IIaLL2"=tyy||b'8A'=bbbb	
 	
 	
 
2IIaLL MU11 6!55 
 MU00 5!44 	 ==??aGMMOOq$8$8L.Q7=EM3Qhh    " Q	 } 333  "!V!M( ( ( $8B55L"*(<";";a"? M : ::=PP  M : ::=PP 
 ?22MMOO66))++88   ))++88   
           L.Q7=EM3Qee  
 L"9w{{}}'9~~~~~   Q1$$LLOOq((LLOOq((LLOOq(( ))++G0E0E0G0G^^          	 	 	 (3J;tyy||TYYq\\DKXXXXr8   c                 `    t          | ||||d           |                     | j                  S NT)r  r!  r   r{   r   r%  rc  ra  s         r6   meta_scatter_reduce_twor  8  s3     dCVTJJJJ>>$*%%%r8   c                 0    t          | ||||d           | S r  r%  r  s         r6   meta_scatter_reduce__twor  ?  s"    dCVTJJJJKr8   c                d    t          j        d                                 cxk     odk    nc  fd                                            dk    r&t          j        |t           j         j                  S t          j                             d          |t           j         j                  S )Nr   r   c                  2    d                                   S )NzAThe probability distributions dimensions must be 1 or 2, but got r   r  s   r6   r[   z"meta_multinomial.<locals>.<lambda>J  s    aTYT]T]T_T_aa r8   r   r  )rL   r]   r{   r   r   rx   r   )r   num_samplesreplacementr   s   `   r6   meta_multinomialr  E  s     
L	EIIKK1aaaa   yy{{a{;ejNNNN;

1{%*U\   r8   c                     d}| D ]}||z  }|S r*  r<   )vsrY  vs      r6   multiply_integersr  S  s&    	A  	QHr8   c                     t          j        t                    k    fd           dz   t          j        t                     k     fd           t          j        t          d  dd          D                       ot          d D                        fd            d d         \  }}||gR S )Nc                  .    d  dt                     S )Nz%It is expected output_size equals to , but got size r  )num_spatial_dimsrm  s   r6   r[   z'upsample_common_check.<locals>.<lambda>]  s#    k8HkkY\]hYiYikk r8   r   c                  .    d  dt                     S )Nz$It is expected input_size equals to r  r  )expected_input_dimsr@  s   r6   r[   z'upsample_common_check.<locals>.<lambda>b  s#    l7Jll[^_i[j[jll r8   c              3   "   K   | ]
}|d k    V  dS r   Nr<   rB   r  s     r6   ri   z(upsample_common_check.<locals>.<genexpr>f  s&      **aAE******r8   c              3   "   K   | ]
}|d k    V  dS r  r<   r   s     r6   ri   z(upsample_common_check.<locals>.<genexpr>f  s&      2N2NQ1q52N2N2N2N2N2Nr8   c                      d  d S )NzDInput and output sizes should be greater than 0, but got input size z and output size r<   )r@  rm  s   r6   r[   z'upsample_common_check.<locals>.<lambda>g  s)     A A A3>A A r8   )rL   r]   r   r  )r@  rm  r  r  channelsr  s   ```  @r6   upsample_common_checkr  Z  s   	LK,,kkkkk   +Q.	LJ..lllll  
 
L**:abb>*****Ns2N2N+2N2N2N/N/N	A 	A 	A 	A 	A   ""1"~FHH+{+++r8   c                 p    t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |                              t          j	                             S )Nr   r   c                  2    d                                   S )Nz>Non-empty 3D data tensor expected but got a tensor with sizes r   r  s   r6   r[   z$upsample_nearest1d.<locals>.<lambda>u      _QVQ[Q[Q]Q]__ r8   r  r   
rL   r]   r   r  r   r  r   r"  rF   r   )r   rm  scalesfull_output_sizes   `   r6   upsample_nearest1dr  o       
LA/

QRR0@AA____   -

kA   ??+,,//1%88 0   r8   c                     t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |          }t          j                   } j	        \  }}}} j
        j        dk    r|dk     rt           j        }|                    |          }|S )	Nr   r   c                  2    d                                   S Nz>Non-empty 4D data tensor expected but got a tensor with sizes r   r  s   r6   r[   z$upsample_nearest2d.<locals>.<lambda>  r  r8   r   r  r   r  r   )rL   r]   r   r  r   r  r   rF   r   r   rx   ro   r   
contiguous)	r   rm  scales_hscales_wr  r   r   rI   
n_channelss	   `        r6   upsample_nearest2dr    s     
LA/

QRR0@AA____   -

kA   __-..F /66M  +Az1a|F""zA~~/];;FMr8   rm  r@  r  r  c                 v    t          ||d          t          j         j        dk     fd           t	          d          D ]9t          j                                                k     fd           :                     |                              t          j	                             S )Nr   r  r  c                      d j          S NzFExpected grad_output to be a tensor of dimension 4 but got: dimension r  r  s   r6   r[   z-upsample_nearest2d_backward.<locals>.<lambda>      kYdYikk r8   c            
      R    d d           d d                                S )NzCExpected grad_output to have the same shape as output; output.size() = z but got grad_output.size(r   r  r  r   s   r6   r[   z-upsample_nearest2d_backward.<locals>.<lambda>  s]    J !J J'7':J J-.J J4?4D4DQ4G4GJ J r8   r   )
r  rL   r]   r   r   r   r   r"  rF   r   )r  rm  r@  r  r  r  r   s   `    @@r6   upsample_nearest2d_backwardr    s     -K!   
LAkkkk   1XX 
 
Q#3A#66     	
 	
 	
 	
   ,,//1+>> 0   r8   c                 p    t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |                              t          j	                             S )Nr   r   c                  2    d                                   S )Nz>Non-empty 5D data tensor expected but got a tensor with sizes r   r  s   r6   r[   z$upsample_nearest3d.<locals>.<lambda>  r  r8   r.   r  r   r	  )r   rm  scales_dr  r  r  s   `     r6   upsample_nearest3dr!    r  r8   c                    t          j        |           t          j        | t           j                  }}||t          |t                    sJ t          |t                    sJ |j        }|                                }	t          ||          }t          ||          }|                    ||	           |                    ||	           t          ||           t          ||           ||fS ||fS )Nr   )r  r  )
rL   r   r   rf   r   r   r   r!   r   r#   )
r   stabler{   
descendingr   r   r  r   r   
out_strides
             r6   	meta_sortr&    s     D!!5#3D#L#L#LqAg1&*-----':..... G	XXZZ
"6955#GY779j111Iz222F3333G4444wa4Kr8   c           	          t          j         j        dk     fd           t          j         j        j        k     fd                                d          wt          j        j        dk    fd           t          j                                        k    fd           t          j        j        j        k    fd           t          j        j        dk    fd                                d	          z  z  t          j                                        k     fd
           t          j        t           fdfD                       d            d S )Nr   c                       j          dS Nz != 2r  )input_gatess   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>      ;3C0J0J0J r8   c                  &    j          d j          S N != r2  )hidden_gatesr*  s   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  s    ;$>>,*<>> r8   r   c                       j          dS )Nz != 1r  )
input_biass   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  s    jo3L3L3L r8   c                  6                                      d  S r-  r  )
gates_sizer1  s   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  s!    z''));;z;; r8   c                  &    j          d j          S r-  r2  )hidden_biasr1  s   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  s    z'@@[->@@ r8   c                       j          dS r)  r  )prev_hiddens   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  r+  r8   r   c            
      p                                      d                    d           d d d  d
S )Nr.  r   z * z // z (aka rn   )r   r   )expected_prev_hidden_numelfactorr3  r*  r7  s   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>  sL    ;$$&&K,<,<Q,?,?JTZb| r8   c              3   8   K   | ]}|j         j         k    V  d S r3   r  )rB   rC   r*  s     r6   ri   z&rnn_cell_checkSizes.<locals>.<genexpr>  sA       
 
 H**
 
 
 
 
 
r8   c                      dS )Nz%expected all inputs to be same devicer<   r<   r8   r6   r[   z%rnn_cell_checkSizes.<locals>.<lambda>
  s    7 r8   )rL   r]   r   r   r   r   r  )r*  r/  r1  r5  r:  r7  r9  r3  s   ``````@@r6   rnn_cell_checkSizesr=    s    
L!Q&(J(J(J(JKKK	L\//>>>>>   !!!$$JZ_)+L+L+L+LMMM*,;;;;;	
 	
 	
 	 11@@@@@	
 	
 	
 
L!Q&(J(J(J(JKKK!,!1!1!!4!4z!AV!K	L99   
L 
 
 
 
"J[I
 
 
 	
 	
 	87    r8   c                     t          | |||d|           t          j        | t          j                  }t          j        |t          j                  }t          j        |t          j                  }|||fS )Nr  r   )r=  rL   r   r   )r*  r/  cxr1  r5  	workspacehycys           r6   _thnn_fused_lstm_cell_metarC    sq     \:{ArRRR E<STTTI		"E,C	D	D	DB		"E,C	D	D	DBIr8   c                 t   t          |          dk    }|r%t          |          }|d         }| j        d         }n:|
r| j        d         n| j        d         }|
r| j        d         n| j        d         }d}|rdnd}|dk    r|n|}|r|||z  g}n|
r||||z  gn||||z  g}|                     |          }|	|z  ||g}|t          j        d| j                  }n|                    |          }|                    |	|z  ||g          }|rdnd}|                     |t          j                  }|||||fS )Nr   r   r   r   r  r   )r   r   r   rL   r   rx   r  )r   r5  weight_stride0
weight_bufhxr?  rU  hidden_size	proj_size
num_layersbatch_firstdropouttrainbidirectionalbatch_sizesdropout_stateis_input_packed
seq_length
mini_batchbatch_sizes_sumnum_directionsout_sizer   r   
cell_shaperB  rA  reserve_shapereserves                                r6   
_cudnn_rnnrZ    s   & +&&!+O %%
 ^
+a.'2FU[^^A
'2FU[^^A
'.QQQN%NNyyH 
$h&?@		 EZX%>??j(^*CD 	
 __Y''F~-z;GJ	z[5<000\\*%%	zN2JI	J	JB %AAAMoom5;o??G2r7J..r8   c                    |r| j         d         n| j         d         }|r| j         d         n| j         d         }|
}|r|||gn|||g}|                     |          }|t          j        d| j                  }n|                    |j                   }|t          j        d| j                  }n|                    |j                   }t          j        d| j        t          j                  }||||fS )Nr   r   r  r   )r   r   rL   r   rx   r  )r   w0w1w2w3hx_cx_r   rO  rU  rH  rJ  
has_biasesrN  rK  rM  rR  rS  output_chanelsr   r   rA  rB  r@  s                           r6   mkldnn_rnn_layerrd  U  s    & $/BQEKNJ#.BQEKNJ N 	6Z00*n5 
 __Y''F
{[5<000]]39%%
{[5<000]]39%%Ael%+FFFI2r9$$r8   c                     | j         dk    r%t          j        dk    pdk    fd           d S t          j        |                               dk    fd           d S )Nr   r   c                       d  S )Nz4: Expected reduction dim -1 or 0 for scalar but got r<   r{   r  s   r6   r[   z'zero_numel_check_dims.<locals>.<lambda>  s    wYYTWYY r8   c                       d  dS )Nz: Expected reduction dim z to have non-zero size.r<   rg  s   r6   r[   z'zero_numel_check_dims.<locals>.<lambda>  s    wUUUUU r8   )r   rL   r   r   )r   r{   r  s    ``r6   zero_numel_check_dimsri  }  s    yA~~1H!r	YYYYY	
 	
 	
 	
 	

 	IIcNNaUUUUU	
 	
 	
 	
 	
r8   c                      |5t          ||                                          }t          ||            d S t          j        |                                dk     fd           d S )Nr   c                        dS )Nz@: Expected reduction dim to be specified for input.numel() == 0.r<   r  s   r6   r[   z%check_argmax_argmin.<locals>.<lambda>  s    t]]] r8   )r   r{   ri  rL   r]   r   )r  r   r{   s   `  r6   check_argmax_argminrl    sq    
S$((**--dC.....JJLLA]]]]	
 	
 	
 	
 	
r8   c                     t          d| |           t          j        | j        ||fnd           }t	          | ||          }|                     |t          j                  S )Nargmaxr   )rl  rF   r  r   r  r   rL   r   )r   r{   r  r  r   s        r6   argmax_argmin_metaro    s[    $,,,
coSFF4PPD$T499E>>%u{>333r8   c                 l    |t           j        k    rt           j        }t          j        d||||          S )Nr<   r  )rL   jaggedr  r   )r  rT   rw   rx   ry   s        r6   scalar_tensorrr    s<    
 ;
%v*   r8   c                    t          ||                                 d          }|                                 dk    rdn|                     |          }t          j        |           t          j        ||k    d            t          | j                  }t          |          dk    r|||<   | 	                    |          | 	                    |t          j
                  fS )NTr  r   r   c                      dS )Nzk not in range for dimensionr<   r<   r8   r6   r[   ztopk_meta.<locals>.<lambda>  s    )G r8   r   )r   r{   r   rL   r  r]   r   r   r   r   r   )r   r]  r{   largestrM  	sliceSizetopKSizes          r6   	topk_metary    s     dhhjjd
;
;
;CXXZZ1__$))C..I		Li!G!GHHHDJH
8}}q>>(##T^^HEK^%P%PPPr8   c                     ||
J d            |                                 }|                                  }	t          j        ||	j        |	j        |	j                  S )Nz;segment_reduce(): Either lengths or offsets must be defined)rT   rx   rw   )r  rL   r   rT   rx   rw   )
r>  r   rs  rc  rn  ro  rp  rr  data_contiggrad_contigs
             r6   meta__segment_reduce_backwardr}    so    
 '"5"5E #6"55 //##K//##K!!	   r8   c                 4   ddl m} t          |                                 d          |                                 dk    r|                               nd}t          j         ||dk    ||k              fd           t          | j        d          | j        dz   d          z             }|r.|                                 dk    r|	                    d           | 
                    |          | 
                    |t
          j                  fS )Nr   )sym_andTrt  r   c                      d  S )Nz9kthvalue(): selected number k out of range for dimension r<   r   s   r6   r[   zkthvalue_meta.<locals>.<lambda>  s    QCQQ r8   r   )r  r  r   r{   r   rL   r]   r   r   r+  r   r   )r   r]  r{   r  r  dimSizer   s     `    r6   kthvalue_metar    s    >=====
dhhjjd
;
;
;C $

QdiinnnAG	LQW%%QQQQ  
 DSD!DJsQwyy$99::E 488::>>S!>>%  $..ek."J"JJJr8   c                    | | n|}t          j        |                                dk    d            |                                }| ,t          j        |                                 |k    d            |,t          j        |                                |k    d            t          j        |                                |k    d            t          j        |                                |k    d            t          j        |                                dk    d            t          j        |                                |d         |d	         z  d
z  k    d            d S )Nr   c                      dS N r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>  s    " r8   c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>       r8   c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>  r  r8   c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>       r8   c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>  r  r8   c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>  s    r r8   r   r   r  c                      dS r  r<   r<   r8   r6   r[   z(checkLSTMBackwardSizes.<locals>.<lambda>  s    R r8   )rL   r]   r{   r   r   )grad_hygrad_cyr?  rB  r@  defined_gradexp_sizes          r6   checkLSTMBackwardSizesr    s2   %177wL	L!!##q(**555  ""HW\\^^x/<<<W\\^^x/<<<	Lh&

333	Lh&

333	LA%zz222	L""hqkHQK&?!&CCZZPPPPPr8   c                     | |dS t          | ||||           t          j        |t                    }t          j        |t                    }|r|                    dd          nd }|||fS )NNNNr   r   F)r  )r  rL   r   legacy_contiguous_memory_formatr  )	r  r  r?  rB  r@  has_bias
grad_gatesgrad_cxr  s	            r6   #_thnn_fused_lstm_cell_backward_implr    s    7?7GRY???!!@  J r1PQQQG4<F
q%000$Iw	))r8   c                 `   d }d }d }|d         r'|                     |                                           }|d         s|d         re|                     |                    d          |                     d          f          }|                     |                    d                    }|||fS )Nr   r   r   r   r  )r  r  r  r  r  grad_weightr  s          r6   linear_backwardr    s    JKI1~ ;!++FKKMM::
1~ BQ B",,l.?.?.C.CV[[QS__-UVV **<+<+<R+@+@AA	Y//r8   c                     t           j                  dk    r j        d         ||z  z  dk    sJ d j         d|             d  fd} j        d         ||z  z  } j        d         |z  } j        d	         |z  }g  j        d d         |||R }                     |          }|                     |            
          }|S )Nr   r$  r   z'Invalid input shape for pixel_shuffle: z with upscale_factor = c                 \    t           j                            |           t           j        k    S r3   r  r  s    r6   r  z,meta_pixel_shuffle.<locals>.is_channels_last  s!    "88==ATTTr8   c                  $               r+t                    dk    rt          j        S t          j        S                     t          j                  rt          j        S                     t          j                  rt          j        S d S r  )r   rL   r   r  rZ  r  )r  r   s   r6   r  z.meta_pixel_shuffle.<locals>.pick_memory_format  s    D!! 	)4  F**..**e.EFF 	)**e.CDD 	)((	) 	)r8   r  r   r   )r   r   r   r"  )	r   upscale_factorr  r  HrWrr   r   r  s	   `       @r6   meta_pixel_shuffler  
  s    	DJ!
2.>2Q RVW W W We$*eeUcee !X WW
U U U	) 	) 	) 	) 	) 	) 	
2>N:;A	B.	(B	B.	(B-$*SbS/-1-b-"--I
..
#
#C
&&1133&
4
4CJr8   c                 L   |                      | j                  }|                     |j                  }|                     |j                  }|                     |j                  }|                     |j                  }|                     |j                  }|||||||fS r3   r  )r   weight0weight1weight2weight3r`  cx_tmpr   hy_cy_grad_output_r_optgrad_hy_r_optgrad_cy_r_optr   rU  rH  rJ  rb  rM  rN  rO  rK  r@  diff_xdiff_hxdiff_cxdiff_w1diff_w2diff_bs                                r6   mkldnn_rnn_layer_backwardr  *  s    4 __U[))FmmCI&&Gv|,,G..G..Gw}--F7GVVWgEEr8   )	out_int32r   c                t    t          j        | |rt           j        nt           j        t           j                  S )NrT   r   )rL   r   r=  r   r   )r   
boundariesr  r   s       r6   meta_bucketizer  M  s7     &7ekkEK-   r8   d   c                     dt                     dk    r*t          j                                          fd           t                     dk    r(                                 rt	          j        d           t          j        t          t                    fd           t          j        dk    fd           t          j        t          t                    fd	           t          j        t          t                    fd
           t          j        k    d            t          j	         j
         j                  S )Nzhistc()r  c                      d j          dS )Nz%"histogram_cpu" not implemented for 'r  r   r  s   r6   r[   zmeta_histc.<locals>.<lambda>^  s    LekLLL r8   r   z%_histc_cuda with floating point inputc                  ,     dt                      S )Nz#: argument 'bins' must be int, not r  binsr  s   r6   r[   zmeta_histc.<locals>.<lambda>d  s    7KKtDzzKK r8   r   c                       d  S )Nz: bins must be > 0, but got r<   r  s   r6   r[   zmeta_histc.<locals>.<lambda>f  s    g#Q#Q4#Q#Q r8   c                  ,      dt                     S )Nz%: argument 'min' must be Number, not r  )r  r   s   r6   r[   zmeta_histc.<locals>.<lambda>i      7LLcLL r8   c                  ,      dt                     S )Nz%: argument 'max' must be Number, not r  )r  r  s   r6   r[   zmeta_histc.<locals>.<lambda>m  r  r8   c                      dS )Nz&{fn_name}: max must be larger than minr<   r<   r8   r6   r[   zmeta_histc.<locals>.<lambda>o  s    %M r8   r   )r   rL   r]   r   rF   r  rf   r   r   r   rx   rT   )r   r  r   r  r  s   ````@r6   
meta_histcr  W  sv    G5U""##%%LLLL	
 	
 	
 5V##(?(?(A(A#%&MNNN	L4!!KKKKK   
LQQQQQRRR	L3LLLLL   
L3LLLLL   
LMMNNN;tELDDDDr8   c                     t                                           |d          }t          j                                         dk    p2t          d                                  dd          D                        fd                                |                              t          j	                             S )Nr   r  r   c              3   "   K   | ]
}|d k    V  dS r  r<   )rB   r   s     r6   ri   z,meta_upsample_bimode2d_aa.<locals>.<genexpr>  s&      !H!Ht$(!H!H!H!H!H!Hr8   r   c                  2    d                                   S r  r   r  s   r6   r[   z+meta_upsample_bimode2d_aa.<locals>.<lambda>  r  r8   r   )
r  r   rL   r]   r   r  r   r"  rF   r   )r   rm  r  r  r  r  s   `     r6   meta_upsample_bimode2d_aar  s  s     -

kA   
LHc!H!Huzz||ABB7G!H!H!HHH____   ??+,,//1%88 0   r8   c                 f    t          ||d          t          j         j        dk     fd           t	          d          D ]1t          j         j                          k     fd           2                     |                              t          j	                             S )Nr   r  r  c                      d j          S r  r  r  s   r6   r[   z4meta_upsample_bimode2d_aa_backward.<locals>.<lambda>  r  r8   c            
      R    d d           d d                                S )NzD
Expected grad_output to have the same shape as output; output.size(r  z
but got grad_output_size(r   r  s   r6   r[   z4meta_upsample_bimode2d_aa_backward.<locals>.<lambda>  sR     9DE9 9K[\]K^9 99 9!,!1!1!!4!49 9 r8   r   )
r  rL   r]   r   r   r   r   r"  rF   r   )r  rm  r@  r  r  r  r  r   s   `     @@r6   "meta_upsample_bimode2d_aa_backwardr    s     -K!   
LAkkkk   1XX 
 
a $4Q$779 9 9 9 9 9	
 	
 	
 	
   ,,//1+>> 0   r8   c                 6   t          j        |                                dk    d            t          j        |                                dk    d            t          j        |j        j        d            t          j        |j        j        d            d S )Nr   c                      dS )Nz%found_inf must be a 1-element tensor.r<   r<   r8   r6   r[   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>      (O r8   c                      dS )Nz%inv_scale must be a 1-element tensor.r<   r<   r8   r6   r[   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r  r8   c                      dS )Nz!found_inf must be a float tensor.r<   r<   r8   r6   r[   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>      3 r8   c                      dS )Nz!inv_scale must be a float tensor.r<   r<   r8   r6   r[   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r  r8   )rL   r]   r   rT   r   )r   r  	inv_scales      r6   *_amp_foreach_non_finite_check_and_unscale_r    s    	LQ O O   
LQ O O   
L)33   
L)33    r8   c                 *    t          j        |           S r3   ri  )r   nanposinfneginfs       r6   
nan_to_numr    rk  r8   c                     | j         t          j        t          j        t          j        t          j        hvsJ d| j          d            | j        }t          ||          }t          ||          }||k    r| S t          | 	                                          }t          | 
                                          }||         ||         c||<   ||<   ||         ||         c||<   ||<   |                     ||           | S )Nz>torch.transpose_: in-place transposition is not supported for z layout)rw   rL   r  
sparse_cscr  
sparse_bscr   r   r   r   r   r   )r   dim0rY  ndimsr   r   s         r6   r  r    s   ;	    	^]]]   IE$&&D$&&Dt||		D$++--  F!'vd|F4L&,!$ZdDJT
T6"""Kr8   c                 (   | j         }| j        rF|                                 }|                                 }|dk    r|dk    sJ d| d| d            n&|                                 dk    sJ d| d            t          | d|dk     rdnd          S )	Nr   r   zEt_ expects a tensor with <= 2 sparse and 0 dense dimensions, but got z sparse and z dense dimensionsz6t_ expects a tensor with <= 2 dimensions, but self is r  r   )r   r  r  r  r{   r  )r   r  r  r  s       r6   t_r    s    IE~ 

__&&
NN$$	Q9>>>L!L L/8L L L $2>1>
 xxzzQMUMMM  dAEAIIqq1555r8   )r  r   sidesorterc                    t          j        t           j                  dk    p j        d d         j        d d         k     fd           t          j        d u p j        j        k     fd           t          j        |dk    p| d           |rt           j        nt           j        }t          t           j                  r!t          j        |t           j	                  S t          j
        d| j        	          S )
Nr   r   c                  \    dt          j                   dt           j                   S )Nztorch.searchsorted(): boundaries tensor should be 1 dimension or the first N-1 dimensions of boundaries tensor and input value tensor must match, but we got boundaries tensor z and input value tensor r   r   )r   sorted_sequences   r6   r[   z#meta_searchsorted.<locals>.<lambda>  s>    5378M3N3N5 5 #'tz"2"25 5 r8   c                  d    dt           j                   dt          j                  ng  S )Nz[torch.searchsorted(): boundary and sorter must have the same size, but got boundary tensor z and got sorter tensor r  )r  r  s   r6   r[   z#meta_searchsorted.<locals>.<lambda>
  sK    B#'(=#>#>B B%+%7tFL!!!RB B r8   r   zetorch.searchsorted(): side and right can't be set to opposites, got side of left while right was Truer  r<   r  )rL   r]   r   r   r=  r   rf   r   r   r   r   rx   )r  r   r  r   r  r  rT   s   ``   ` r6   meta_searchsortedr    sA    
LO!""a' 	9 "%CRC8	
 	
 	
 	
 	
	 	 	 
L$?//6<?	
 	
 	
 	
 	
   
L#e)	$   %5EKK%+E$%% KU-D
 
 
 	
 {2U?3IJJJJr8   c                 ~     t          j         t           j        t           j        t           j        fv fd           d S )Nc                      d  S )Nz/Unsupported input type encountered for isin(): r<   r   s   r6   r[   z3_check_for_unsupported_isin_dtype.<locals>.<lambda>$  s    I%II r8   )rL   r]   r  
complex128	complex64r   s   `r6   !_check_for_unsupported_isin_dtyper  !  sC    	Lej%"2EODDIIII    r8   c                 Z    |                      ||                     d          f          }|S )Nr   r  )r  r   num_weightsr  r  r  s         r6   meta_embedding_dense_backwardr  (  s0     ''k6F6Fr6J6J(KLLKr8   c                 |    |	r#t                               | ||||||||
|
  
        S t          | ||||||||
|
  
        S r3   )r+   _embedding_bag_sparse_backward!meta_embedding_bag_dense_backward)r>  r   ro  r  r  maximum_indicesr  r  rU  r  r  r  s               r6   meta_embedding_bag_backwardr  4  st      
22
 
 	
 1
 
 	
r8   c
                 4    t          j         j        t           j        t           j        t           j        t           j        fv  fd           |t          k    rt          j        |d u                                | 	                    d          f          }
|
S )Nc                      d j          S )Nz$Unsupported input type encountered: r   )r>  s   r6   r[   z3meta_embedding_bag_dense_backward.<locals>.<lambda>n  s    CtzCC r8   r   )
rL   r]   rT   rW  rX  rV  float64r  r   r   )r>  r   r  r  r  r  r  rU  r  r  index_grad_weights   `          r6   r  r  _  s     
L
u}enemU]SSCCCC   x_D0111TYYq\\'BCCr8   c                    |                      d          }t          j        |t          k    d           t          j        |                                 dk               t          j        |                                dk               |                     d          }t          j        |                                dk               t          j        |                     d          |k               |                     |f          }	|	S )Nr   zHembedding_bag_backward: per_sample_weights only supported for mode='sum'r   r   )r   rL   r]   r  r{   r   )
r>  r5  r   ro  r  rU  r  embedding_featuresr  r   s
             r6   .meta_embedding_bag_per_sample_weights_backwardr 	  v  s     1	LR   
Lq!!!	L!#$$$,,q//K	L"###	LQ#55666^^[N++FMr8   )assume_uniqueinvertc                   t          j        t          | t                    pt          |t                    d            t          | t                    st          j        | |j                  } t          |t                    st          j        || j                  }t          | j                   t          |j                   t          j        | t           j	                  S )Nc                      dS )Nz<At least one of elements and test_elements must be a Tensor.r<   r<   r8   r6   r[   zmeta_isin.<locals>.<lambda>  r  r8   r  r   )
rL   r]   rf   r   r,  rx   r  rT   r   r  )elementstest_elementsr	  r	  s       r6   	meta_isinr	    s     
L8V$$I
=&(I(INN   h'' G<1EFFFmV,, L]8?KKK%hn555%m&9:::HEJ7777r8   r  c                     t          j        | dk    d            t          |t          j                  \  }}t          j        ||          S )Nr   c                      dS )Nz,polygamma(n, x) does not support negative n.r<   r<   r8   r6   r[   z meta_polygamma.<locals>.<lambda>  s    !O r8   r  r   )rL   r]   r   r   r  r   )r  r   rI   rD   s       r6   meta_polygammar
	    sW     
LaOOPPP(;H  OA| D5555r8   c                      t          d          )Nz.Tensor.item() cannot be called on meta tensors)r!  r   s    r6   meta_local_scalar_denser	    s    
G
H
HHr8   c                 *    t          j        |           S r3   ri  r   s    r6   silur	    rk  r8   c                 j    t          | t          j                  \  }}t          j        | |          S r  )r   r   r  rL   r   )r   rI   rD   s      r6   sigmoidr	    s=     );H  OA| D5555r8   c                    |                                  dk    }|                                 dk    }|r|r@|                    d          |                     d          |                    d          g}nVt          j        |                    d          |                    d          k    d            |                     d          |                    d          g}n|rkt          j        |                    d          |                     d          k    d            |                     d          |                    d          g}n~t          j        |                     d          |                    d          k    d            |                     d          |                     d          |                    d          g}|p| j        }t          j        j        rUd|j        z  }|d         |z   dz
  |z  |z  }||k    r|d         |z  |dg}	n|dg}	t          j        ||	|| j	        	          }
nt          j
        ||| j	        	          }
|
S )
Nr   r   r   c                      dS Nz matrix batch sizes have to matchr<   r<   r8   r6   r[   z2_create_grouped_mm_output_tensor.<locals>.<lambda>      6X r8   r   c                      dS r	  r<   r<   r8   r6   r[   z2_create_grouped_mm_output_tensor.<locals>.<lambda>  r	  r8   c                      dS )Nzbatched dimension has to matchr<   r<   r8   r6   r[   z2_create_grouped_mm_output_tensor.<locals>.<lambda>  s    6V r8   rU  r  )r{   r   rL   r]   rT   r7  r   itemsizer!  rx   r   )rB  rD  offsr9  
mat1_is_2d
mat2_is_2drV  	alignmentsize_paddedr%  r   s              r6    _create_grouped_mm_output_tensorr	    s1   qJqJ C 	5		!diillDIIaLLAHHL		!		!,.X.X   		!diimm4HH 
	CL		!		!,.X.X   		!diill3HH L		!		!,.V.V   		!diillDIIbMMBH'TZI} I),,	|i/!3	AIM##"1+3[!DJJ%q)J!j	$+
 
 
 k()DKHHHJr8   mat_amat_br	  c	                 l    t          j        d u d u k    d            d uod u}	|	rWt           j        j        rt           j        nt           j        }
t          j         j        |
k    o
j        |
k     fd           nAt          j         j        t           j        k    oj        t           j        k     fd           t          j                                         dv o                                dv  fd                                            dk    }                                dk    }|r|s?t          j         	                    d          	                    d          k    d	           |	rHd
 }d }t          j         |            fd           t          j         |          fd           d } |d             |d           t          j        j        t           j
        k    rj        t           j
        k    p)j        t           j        k    oj        t           j        k    fd           j        t           j        k    oj        t           j        k    d dfd	}|r|rj        d         nd} |d d|            |dd|           t          j        |d u d            |s|rwt          j        d u fd           Yt          j                                        dk    fd           t          j        j        t           j        k    fd           nt          j        d u d            t          j        |d u d            t          j        |d u p|t           j        k    d            t           |          S ) Nc                      dS )Nz,Either both scale factors are given, or noner<   r<   r8   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    > r8   c                  *    d j          dj          dS )Nz5Expected inputs of E4M3 FP8 type but got mat_a.dtype= and mat_b.dtype=r  r   r	  r	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s!    xEKxxjojuxxx r8   c                  *    d j          dj          dS )Nz1Expected inputs of BF16 type but got mat_a.dtype=r#	  r  r   r$	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s!    tttfkfqttt r8   )r   r.   c                  \    d                                   d                                  S )Nz3Multiplicands must be 2D or 3D but got mat_a.dim()=z and mat_b.dim()=r   r$	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s-    qeiikkqqdidmdmdodoqq r8   r   r   r  z3contraction dimension of mat_a and mat_b must matchc                 Z    |                                  }|d         dk    o|d         dk    S Nr  r   r   r  mat
mat_strides     r6   r  z-_meta_grouped_mm_common.<locals>.is_row_major  s,    Jb>A%=*R.A*==r8   c                 Z    |                                  }|d         dk    o|d         dk    S r(	  r  r)	  s     r6   r  z-_meta_grouped_mm_common.<locals>.is_col_major  s,    Jb>Q&=:b>A+==r8   c                  B    d                                  dd           S )NzNExpected mat_a tensor to be row major in the last two dimensions, got strides r  r  )r	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>%  s)    zejeqeqesestvtwtwexzz r8   c                  B    d                                  dd           S )NzQExpected mat_b tensor to be column major in the last two dimensions, got strides r  r  )r	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>)  s)    }hmhththvhvwywzwzh{}} r8   c                 D                                     dz
  d                                z  }                                dz
           dk    rQ         t          dj        dz
                     k    r)t          j                 |z  dk     fd           d S          dk    rTdz
           t          dj                           k    r,t          j        dz
           |z  dk     fd           d S t          j        dfd           d S )Nr   rU  r   c                  (    d d  d           dS )Nr   stride along % dim to be multiple of 16 bytes, got r  r<   end_dimmat_namer+	  s   r6   r[   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda>5  sB      AH  A  AG  A  Ajtu|j}  A  A  A r8   c                  4    d d dz
   d dz
            dS )Nr  r1	  r   r2	  r  r<   r3	  s   r6   r[   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda><  s\      IH  I  IGaK  I  Inx  zA  DE  zE  oF  I  I  I r8   Fc                       d d j          dS )NzInvalid strides/sizes, got z for strides and z for sizes.r2  r)	  s   r6   r[   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda>A  s    ijiiSVS\iii r8   )r{   element_sizer   r  r   rL   r]   )r5	  r*	  r	  r4	  r+	  s   `` @@r6   check_valid_stridesz4_meta_grouped_mm_common.<locals>.check_valid_strides,  s   ''))a-#**,,,	ZZ\\
gk"a''Jw,?3sy1%D
 D
 -
 -
 L7#i/14 A  A  A  A  A  A      A%%*Wq[*ASsy!F
 F
 +
 +
 L7Q;')3q8 I  I  I  I  I  I    
 Liiiii    r8   r	  r	  c                  *    d j          dj          dS )NzhFor FP8 scales must both be float32, or for MXFP8 both scales must be float8_e8m0fnu. Got scale_a.dtype=z and scale_b.dtype=r  r   r  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>N  sE      r  @G  @M  r  r  bi  bo  r  r  r r8   c                     | |z   dz
  |z  |z  S )z$Rounds up x to nearest multiple of yr   r<   rC   r=  s     r6   round_upz)_meta_grouped_mm_common.<locals>.round_upU  s    UQY1$))r8   r   c                    	
                                  dk    rt          j                                         fd           rDt          j                                                                          k     fd           d S t          j                                         dk     fd           t          j        j        d         j                 z  k     fd           d S t          j                            d          dk     fd	           t          j        j        d         j        d         k     fd
           rt          j        j        j        k     fd           j        \  }}d} ||z  d          	 |d          
t          j        j        d         	k    oj        d         
k    	
fd           d S t          j                                         dk     fd           t          j        j        d         j        dz            k     fd           d S )Nr   c                      d  dS )Nr  z to be contiguous.r<   
scale_names   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>]  s    F
FFF r8   c                  .    d d j          dj          S )NzKFor MXFP8, scale must have same number of dimensions as target tensor, but  has mat.ndim= and scale.ndim=r  r*	  rs  rA	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>f  sN      !onx  !o  !o  IL  IQ  !o  !o  ch  cm  !o  !o r8   r   c                  :    d d                                   dS )Nr  z to be 1D tensor, but got 	D tensor.r   rs  rA	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>k  (     hJ h hRWR[R[R]R] h h h r8   r   c                  N    d d j                  z   dj         d          dS )Nr  z	 to have r  r   z
 elements.r2  )r*	  rs  scale_multiplierrA	  
scaled_dims   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>o  s^      !UJ  !U  !U:AVYiAi  !U  !Uz  {F  GH  {I  !U  !U  !U r8   r   c                      d  dS )Nr  z( to be contiguous in the last dimension.r<   r@	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>t  s    \
\\\ r8   c                  H    d d j         d          dj         d          dS )Nr  z batch dimension to be r   , got r  r2  rE	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>x  s4    p
pp39UV<pp_d_jkl_mppp r8   c                  .    d d j          dj          S )NzMFor MXFP8, scale should have same number of dimensions as target tensor, but rC	  rD	  r  rE	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  sN      !qpz  !q  !q  KN  KS  !q  !q  ej  eo  !q  !q r8   r  r  r4  r  c            
      :    dj          d  d d dj          
S )NzFor MXFP8, expected mat.shape=z to have scale shape of (,z), but got r2  )G	blocked_K	blocked_Nr*	  rs  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  sb      !Y  !Y  !Yef  !Y  !Yir  !Y  !Yu~  !Y  !Y  LQ  LW  !Y  !Y r8   c                  :    d d                                   dS )Nr  z to be 2D tensor, but got rG	  r   rH	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  rI	  r8   c                  N    d d j         dz             dj         d          dS )Nr  z non-batch dimension to be r   rO	  r  r2  )r*	  rs  rA	  rL	  s   r6   r[   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  s]      !FJ  !F  !FSVS\]^ak]kSl  !F  !Ftyt  AB  uC  !F  !F  !F r8   )r{   rL   r]   rZ  r   r   r   )rA	  rs  r*	  rL	  rK	  ra  r  rn  rS	  rT	  rU	  is_mxfp8r=	  s   `````   @@@r6   check_scalez,_meta_grouped_mm_common.<locals>.check_scaleY  sM   wwyyA~~''))FFFF    L		swwyy0 o  o  o  o  o  o    
 L		q(hhhhh   LA#)J*?BR*RR U  U  U  U  U  U  U  U    
 LL$$)\\\\   KNcil2pppppp    LEJ. q  q  q  q  q  q   $kGAq!!#J (Z ; ;I (C 0 0IL	"2Qsy}	7Q Y  Y  Y  Y  Y  Y  Y  Y    
 L		q(hhhhh   LA#)A
N*CC F  F  F  F  F  F  F    r8   r   r  r  c                      dS )Nz:Scale result tensor provided, but it is not supported yet.r<   r<   r8   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  rc   r8   c                  ^    d                                   d                                  dS )Nz/Offsets tensor not provided, but is needed for zD/zD multiplicand layouts.r   r$	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s0    yeiikkyyUZU^U^U`U`yyy r8   c                  4    d                                   dS )Nz.Offsets tensor must be 1D, but got offs.dim()=r  r   r	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    VVVV r8   c                      d j          dS )Nz7Offsets tensor must be integer (int32) tensor, but got r  r   r]	  s   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    _RVR\___ r8   c                      dS )NzJOffsets tensor provided, but is not needed for 3D/3D multiplicand layouts.r<   r<   r8   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    ` r8   c                      dS )Nz2Bias tensor provided, but it is not supported yet.r<   r<   r8   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    D r8   c                      dS )Nz4If output dtype provided, it must be torch.bfloat16.r<   r<   r8   r6   r[   z)_meta_grouped_mm_common.<locals>.<lambda>  s    F r8   r  )rL   r]   r7  r8  r  rY  rT   rX  r{   r   rV  r  r   r=  r	  )r	  r	  r  r  r	  r7  r  r9  r  scaled	fp8_dtypemat_a_is_2dmat_b_is_2dr  r  r9	  rY	  rK	  rX	  r=	  s   `````             @@r6   _meta_grouped_mm_commonrf	    s    
L	Dgo.>>   D 8WD%8F  

-2]->WE))EDW	K9$A	)Axxxxx	
 	
 	
 	

 	K5>)KekU^.Kttttt	
 	
 	

 
L		v7%))++"7qqqqq  
 ))++"K))++"K 
k 
JJrNNejjnn,A	
 	
 	

  
	> 	> 	>	> 	> 	> 	Lzzzz	
 	
 	
 	L}}}}	
 	
 	

  0 ''''''w2]em+N0N !55 :MU%99 r  r  r  r  r	
 	
 	
 MU11 6!55 	
	* 	* 	*:	 :	 :	 :	 :	 :	 :	z "-+-+-DJqMMST 	 	Iwq2BCCCIwq2BCCCD PP	
 	
 	

  
k 
yyyyy	
 	
 	
 L

aVVVV   L
ek)____  
 	DL``	
 	
 	

 
LDD  
 
LT8Y%.8FF  
 ,E5$	JJJr8   c           
      0    t          | |d d ||d |          S )N)r  r  r	  r7  r  r9  rf	  )r	  r	  r	  r7  r9  s        r6   meta_grouped_mmri	    s3     #	 	 	 	r8   c	                 2    t          | ||||||||	  	        S )N)r  r  r	  r7  r  r9  r  rh	  )	r	  r	  r  r  r	  r7  r  r9  r  s	            r6   meta_scaled_grouped_mmrk	    s6     #!%
 
 
 
r8   rC   half_to_floatc                     |r| j         t          j        k    sJ t          j        | t          j        j                  \  }}|s|n|}t          j        | |t          j                  }|S )Nr  r  )	rT   rL   rN   rF   r   r   rG   r   r   )rC   r{   rl	  computation_dtyperD   r  s         r6   softmaxro	    sy      %w%*$$$$&+&>	uDL' ' '#| (5K<<:KL

1L@W
X
X
XCJr8   c           	         	
 t          j        t                    dz  dk    fd           | j        t                    
t                    dz  }
|z
  	t          j        
|k    
fd           t	          d D                       r| }t          	
          D ]d
z
  dz
  z           dk     r2|                              |j                          z             }dz            dk     r.|                    d|j                 dz            z             }|                                S t          d 	                   }t          |          D ]ot                    dz   dz  z
  	z                     z   dz            z   }t          j        |dk    	fd           |	                    |           pt          j
        || j        | j        | j        t          |                     S )	Nr   r   c                  (    dt                      S )Nz1Length of pad must be even but instead it equals r  rH  s   r6   r[   z'_constant_pad_nd_meta.<locals>.<lambda>  s    NCHHNN r8   c                  0    dt                     d  dS )Nz`Length of pad should be no more than twice the number of dimensions of the input. Pad length is z while the input has z dimensions.r  )l_inprI  s   r6   r[   z'_constant_pad_nd_meta.<locals>.<lambda>  s/     25c((    r8   c              3   V   K   | ]$}t          |t          j                  o|d k    V  %dS r  )rf   rF   IntWithoutSymInt)rB   r:  s     r6   ri   z(_constant_pad_nd_meta.<locals>.<genexpr>  s8      
I
I:a/00;Q!V
I
I
I
I
I
Ir8   r   c            	      X    d z             d          ddz             d z    d	S )NzThe input size z, plus negative padding r   r   zG resulted in a negative output size, which is invalid. Check dimension z of your input.r<   )r   r   l_diffrI  pad_idxs   r6   r[   z'_constant_pad_nd_meta.<locals>.<lambda>&  sf     Mk&1*&= M M7|M M"%gk"2M M17!M M M r8   )rT   rx   rz   r   )rL   r]   r   r   r  r   narrowr   r   r   r   rT   rx   rz   r   )r   rI  r  l_padc_input	new_shapenew_dimr   r   rw	  rs	  rx	  s    `     @@@@@r6   _constant_pad_nd_metar~	    s    
LC1NNNN  
 +KEHHMEU]F	L	 	 	 	 	   
I
IS
I
I
III vu%% 	T 	TA519q=)G7|a!..G}gmA&6W&E  7Q;!##!..Aw}Q/?#gPQkBR/RSS}}[&)**I5\\ 	" 	"c((q1uk*fqj)CL83w{;KKqLM M M M M M M M	
 	
 	
 	!!!!;k|)+E22   r8   r  r  r  c                    |                                  dk    s
J d            | j        }|j        }|j        dk    r
|d         f}n)|j        dk    r|d         |d         f}ng ||d         R }| j        }|                     ||          S )Nr   z'weight' must be 2-Dr   r   r   )r{   r   r   rT   r   )	r5  r   r  r  r  weight_shapeindices_shaper   r9  s	            r6   	embeddingr	  5  s     ::<<14<LMM|q&21o%7					"1%|A7		5m5\!_55	IIY777r8   max_lengthspadding_valuec                     t          |          dk    sJ t          |          dk    sJ |d         j        d         dz
  }|d         }||g| j        dd          R }|                     |          S r  )r   r   r   )r   ro  r	  r	  r  rq  r_  s          r6   $meta__jagged_to_padded_dense_forwardr	  M  s     w<<1{q    
aAAAq,6<+,,LL)))r8   c                 ^    t          |           t                      d                         }|S )Nc                 8    t          | t          j                  S r  rJ   r   r  r)  s    r6   _fz)_create_unary_float_meta_func.<locals>._f_  s$      =J
 
 
 	
r8   r=   r$   funcr	  s     r6   _create_unary_float_meta_funcr	  ^  9    4]]
 
 ] 

 Ir8   c                 ^    t          |           t                      d                         }|S )Nc                 :    t          | |t          j                  S r  r	  r<	  s     r6   r	  z*_create_binary_float_meta_func.<locals>._fj  s&      q!@!M
 
 
 	
r8   r	  r	  s     r6   _create_binary_float_meta_funcr	  i  r	  r8   c                      t                      fd            } j         d}||_         t          t          t          |                    |          }|S )Nc                 T     | g|R i |}t          | j        |j                   | S r3   r  )r   rH   r  r   r5   s       r6   _fnz#_register_inplace_meta.<locals>._fn  s;    b''''''
CI666r8   rI   )r   rp   r=   getattrr+   )r5   r	  inplace_names   `  r6   _register_inplace_metar	    sk    
2YY    Y
 k$$$LCL
4-l33
4
4S
9
9CJr8   c                 B    t          j         j        j        k     fd            g}t          t                    rGj        dk    r't          j         j        j        k     fd           |                               t          |dt          j	        iS )Nc                  (    dj          d j          S )Nrz  z for `end`, but got dtype r   )rq   rr   s   r6   r[   zlerp.<locals>.<lambda>  s    T%+TTTT r8   r   c                  (    d j          dj          S )Nrz  z for `weight`, but got dtype r   )rr   r5  s   r6   r[   zlerp.<locals>.<lambda>  s    b%+bbTZT`bb r8   r>   )
rL   r]   rT   rf   r   r   r   rJ   r   rG   )rr   rq   r5  rH   s   ``` r6   lerpr	    s     
Lsy TTTTT   3<D&*%% ;!Lv|+bbbbb   	F	=E  r8   )r  c                <    t          | ||t          j                  S r  r  r   tensor1tensor2r  s       r6   addcmulr	    s(     w0O0W   r8   c                    t          j        t          j        |j                  ot          j        |j                   d            t          | ||t          j                  S )Nc                      dS )N)zFInteger division with addcdiv is no longer supported, and in a future zErelease addcdiv will perform a true division of tensor1 and tensor2. z4The historic addcdiv behavior can be implemented as zA(input + value * torch.trunc(tensor1 / tensor2)).to(input.dtype) zfor integer inputs and as z6(input + value * tensor1 / tensor2) for float inputs. z?The future addcdiv behavior is just the latter implementation: z4(input + value * tensor1 / tensor2), for all dtypes.r<   r<   r8   r6   r[   zaddcdiv.<locals>.<lambda>  s	     	
 r8   r  )rL   r]   rF   r7  rT   rJ   r   rG   r	  s       r6   addcdivr	    sr     
L"7=11 6&w}55	
		
 		
    w0O0W   r8   c                     i } dD ]#}t           |         }|D ]}|| vr||         | |<   $|                                 D ]\  }}t          |t          j        j                  r&t          |t                    sJ  |                    t          j        j	        j
                  |           t          j                            |                                d          r"|t           d         v rt          | d          |j        r|                                dv rd|                                v rt                              ||           d|                                v rt"                              ||           Hd|                                v rt$                              ||           {d	|                                v rt&                              ||           t(                              ||           d S )
N)ru   post_autogradpre_autogradCompositeImplicitAutogradru   z is a CompositeImplicitAutograd op, we shouldn't register meta function for it. Instead, we should let the decomposition run and write meta kernels for the base operators.>   aten::cloneaten::copy_aten::rot90aten::_to_copyaten::empty_stridedaten::constant_pad_ndaten::as_strided_scatterzmkldnn::zmkl::zonednn::zquantized::)r   itemsrf   rL   _opsHigherOrderOperatorr   py_impl_CDispatchKeyr-   %_dispatch_has_kernel_for_dispatch_keyr  r!  is_view2_meta_lib_dont_use_me_use_register_meta_for_mkldnnimpl/_meta_lib_dont_use_me_use_register_meta_for_mkl2_meta_lib_dont_use_me_use_register_meta_for_onednn5_meta_lib_dont_use_me_use_register_meta_for_quantized'_meta_lib_dont_use_me_use_register_meta)activate_meta_tablero   registryopoop_overloadr5   s         r6   activate_metar	    sb    : 9 9-d3 	9 	9C---+3C=#C(	9 /4466 6N 6NR
 k5:#ABB 	+z222226EH0566r:::899 ;
 
 +	N 8@@@"" ; ; ;   A   	N 	 	 [--////BGGUWXXXXK,,....?DD[RTUUUU{//1111BGGUWXXXX+"2"2"4"444EJJ    8<<["MMMMm6N 6Nr8   r   r  r3   )NNNFr   r   r   r  )Tr  )r  )r  T)FF)TT)rW  )FTN)TFF)TF)r   )ra  N)r/   r  )r<   r   r  F)r<   r   FTN)Fr   FNFr   )NF)r   F)r
  r  FN)NNNNN)r   NNr   )NNF)r-  FFN)Nr-  FFN)r-  FNN)Nr-  FNN)r-  FN)FN)FNNNN)NNNF)Nr   FNN)NNNN)r   TT)NNr   N)r  r   r   )r   )NNNNF)r   FF)r-  (  r  collections.abcr   enumr   	functoolsr   typingr   r   r   r	   typing_extensionsr
   rL   torch._prims_commonr  rF   r   r   r   torch._decompr   r   r   r   
torch._opsr   torch._primsr   r   r   r   r   r   r   r   r   r   r   r   r   torch._prims_common.wrappersr    r!   r"   r#   r$   r  r%   r&   torch.fx.experimentalr'   rW  torch.utilsr(   r9   r)   r*   opsr+   libraryLibraryr	  r   r  r  r  r=   rJ   rV   r_   linspacelogspacer  r   taker$  r   r   r   r   cummaxcumminr   r   r   r   r   r  r   _fft_c2cr   r   r   _fft_r2cr   randpermgenerator_outr  r   r	  randintr  r  low_outr  randr  _fft_c2rr  r  r(  r0  
unsqueeze_r4  _sparse_semi_structured_linearr  rT   rA  _sparse_semi_structured_mmrG  _sparse_semi_structured_addmmrK  _cslt_sparse_mmr`  index_reducerg  index_reduce_ri  index_selectrm  segment_reducerz  r  	unary_outr~  r{   r  r   r  r  r  r  r  _assert_asyncr  msgr  _printr  _make_dep_tokenr  r  _functional_sym_constrain_ranger  r  (_functional_sym_constrain_range_for_sizer  _functional_assert_asyncr  r   r  r   r  r  r  r  _linalg_eighr  r  _linalg_eigvalslinalg_eigvalsr  
linalg_eigr  r  r  r  r  r
  r  r#  linalg_inv_exr&  linalg_ldl_factor_exr\   r,  linalg_ldl_solver;  	linalg_lurA  linalg_lu_factor_exrE  linalg_lu_solverN  	lu_unpackrT  r]  	linalg_qrrd  rh  re  _linalg_svdrr  r8  r  r  r  linalg_solve_triangularr  r  r  _linalg_detr  r  r  r  reflection_pad1dr  replication_pad1dr  r  reflection_pad1d_backwardr  replication_pad1d_backwardr  r  reflection_pad2dr  replication_pad2dr  reflection_pad2d_backwardr  replication_pad2d_backwardr  r-  reflection_pad3dr/  replication_pad3dr2  reflection_pad3d_backwardreplication_pad3d_backwardr9  _pdist_forwardrP   r=  _pdist_backwardrC  baddbmmr]  	bernoullir`  
bernoulli_rd  r:  rg  poissonrj  _fused_moving_avg_obs_fq_helperr|  mmr  r  r   r  r  miopen_batch_normr  convolutionr  r	  _has_mkldnnr	  r  _convolution_pointwiser  _linear_pointwiser  has_mklr	  r  _mkl_linearr  r	  r  qconv2d_pointwiseqconv_pointwiser  binaryr  qlinear_pointwiser,  r  binary_tensorr  linear_dynamic_fp16linear_relu_dynamic_fp16r  r	  r  
max_pool2dr   int4mm_packed_weight_cpur  r  
avg_pool2dr4  r9  avg_pool2d_backwardrA  
avg_pool3dr^  avg_pool3d_backwardri  _adaptive_avg_pool2drn  _adaptive_avg_pool3drq  _adaptive_avg_pool2d_backwardr{  _adaptive_avg_pool3d_backwardr  r~  adaptive_max_pool2dr  r  r  adaptive_max_pool3dr  r  r  repeat_interleaver  rg   r  r  r  r   _unsafe_indexr  convolution_backwardr  addbmmr  randint_liker  _fused_adam__fused_adamw_r   _fused_adamr  _int_mmr  _convert_weight_to_int4packr  #_convert_weight_to_int4pack_for_cpur"  _weight_int4pack_mmr-  _weight_int4pack_mm_for_cpur3  r<  r>  rf  _dyn_quant_pack_4bit_weightrr  _dyn_quant_matmul_4bitrx  _weight_int8pack_mmr  _cdist_forwardr  _cdist_backwardr  _embedding_bagr  _embedding_bag_forward_onlyr  r  nansumr  median	nanmedianr  
dim_valuesrU  r   r  logical_not_r  repeatr  zero_r  mul_Scalardiv_logical_and_logical_or_logical_xor_r  add_sub_r  r  subr  rounddecimalsr  r  
__rshift__r  
__lshift__r  zeror  r  r  fillr  relu_r  	_add_relur	  rrelu_with_noiser  rrelu_with_noise_functionalr  rrelu_with_noise_r  	index_put_unsafe_index_putr  masked_fill_r  _masked_scaler  masked_scatter_r#  masked_scatterr%  masked_scatter_backwardr'  
index_put_r)  r6  bmmr9  r;  r?  rD  r(  r)  rS  rh  re  r   max_pool2d_with_indices_backwardrv  max_pool2d_with_indicesrx  fractional_max_pool2dr  max_pool3d_with_indicesr   max_pool3d_with_indices_backwardr  r  r  r  grid_sampler_2d_backwardr  r  r  r  r  onesr  zerosr  select_scatterr  slice_scatterr  r   r  r  gatherr  r  r  r  r  r  scatter_addr#  scatter_add_r&  r  r%  r  rc  value_reducer*  scatter_r,  #_scaled_dot_product_flash_attentionrE  rP  #_scaled_dot_product_cudnn_attentionrZ  0_scaled_dot_product_fused_attention_overrideabler]  ,_scaled_dot_product_flash_attention_backwardrh  +_scaled_dot_product_flash_attention_for_cpurk  4_scaled_dot_product_flash_attention_for_cpu_backwardro  *_scaled_dot_product_attention_math_for_mpsr  '_scaled_dot_product_efficient_attentionr  0_scaled_dot_product_efficient_attention_backwardr  ,_scaled_dot_product_cudnn_attention_backwardr  _flash_attention_forwardr  _flash_attention_backwardr  _efficient_attention_forwardr  _efficient_attention_backwardSymIntr  
_scaled_mmr  scatter_reducetwotwo_outr  scatter_reduce_r  multinomialr  r  r  r  _upsample_nearest_exact1dr  _upsample_nearest_exact2dr  "_upsample_nearest_exact2d_backwardr!  _upsample_nearest_exact3dr   r#  values_stabler&  r=  _thnn_fused_lstm_cellrC  rZ  rd  ri  rl  rn  argminro  rr  topkry  _segment_reduce_backwardr}  kthvaluer  r   r  r  r  r  pixel_shuffler  r  	bucketize
Tensor_outr  histcr  _upsample_bilinear2d_aa_upsample_bicubic2d_aar   _upsample_bilinear2d_aa_backwardr  r  r  r  r  searchsortedr  r  embedding_dense_backwardr  _embedding_bag_backwardr  _embedding_bag_dense_backwardr  *_embedding_bag_per_sample_weights_backwardr 	  isinr	  	polygammar
	  _local_scalar_denser	  r	  r	  r	  rf	  _grouped_mmri	  _scaled_grouped_mmrk	  _softmaxro	  constant_pad_ndr~	  r	  _jagged_to_padded_dense_forwardr	  r	  r	  special_airy_aispecial_bessel_y0special_bessel_y1special_modified_bessel_i0special_modified_bessel_i1special_modified_bessel_k0special_modified_bessel_k1!special_scaled_modified_bessel_k0!special_scaled_modified_bessel_k1special_chebyshev_polynomial_tspecial_chebyshev_polynomial_uspecial_chebyshev_polynomial_vspecial_chebyshev_polynomial_w&special_shifted_chebyshev_polynomial_t&special_shifted_chebyshev_polynomial_u&special_shifted_chebyshev_polynomial_v&special_shifted_chebyshev_polynomial_wspecial_hermite_polynomial_hspecial_hermite_polynomial_hespecial_laguerre_polynomial_lspecial_legendre_polynomial_pr	  r	  r	  r	  lerp_addcmul_addcdiv_torch._refs.nn.functionaltorch._refs.specialr	  r<   r8   r6   <module>r
     s[    $ $ $ $ $ $             5 5 5 5 5 5 5 5 5 5 5 5 ' ' ' ' ' '  # # # # # # + + + + + + + + + +            " ! ! ! ! ! U U U U U U U U                                       < ; ; ; ; ; ; ; 7 7 7 7 7 7 ) ) ) ) ) ) WT]]Yt__y~*/-*?*?PV*W*W ' %a )X
8BF#3"4hr2v6F"FG 
 
 
 
3   (* * *   t}-..
 

=5 5 5  /.5p 	!49=122' '  32' !)4+<+@ABB%' % % % %  CB%$ t%&&I I  '&I 	[$+/4;+>P  Xy!!  "!  !)4+<+@ABBI I  CBI3 3 3l V $s) 4     %t}'89::K K  ;:K $s)     %t}'89::8
 8
  ;:8
v t}*++"& 3 3 3 3 ,+3 t}$%% *
 
 
 
 &%
 $dl&6788
 *     98&  $,"6788 *     98& 	!49=122%)$tPT      32 %t}'89::$Dv $DDI $Dc $DC $D $D $D  ;:$DN tz!""   #"0( ( ( t&''  (' t233
 "%)'+   6
	
 c] $   43B t.//
 (,	 
  $	   0/0 t122 	
'+  
  	 $   32D t#$$ ""'+"-< -<,-<\-< 6
-< F	-<
 $-< -< -< -< -< -< -< %$-<` t ()) 	I 	I 	I
	I		I 	I L		I
 	I 	I 	I 	I 	I *)	I t!)** 	 	 	
			 	 L		
 	 	 	 	 	 +*	 t ())' ' *) ' t"*++
 !% $ $ W  W  W
 W W f	 W
 f W f W  W  W  W  W  W ,+ WF  $("4566   76 tx|     $("4566   76 tx|    tz!""6 6 #"6 tz~( ( (
 t!)**  +* t!%&&  '& t{"##  $# t#+,, ) ) ) ) -,) t'/00, , , 10, t3;<<   =<
 t08996 6 6 :96& t<DEE  FE
 t,011  21
F C       F  #        N (,
 


 !%
 
 
 
$ V S C    
 
F 
$ 
 
 
 
 
"  	   	   C     !)4+<+HIJJ]N++  s T    ,+ KJ" $,d.A.EFGGB B6 B B B  HGB  !!]N++	6 	 	 	 ,+ "!	Q QF Q Q Q Q t*++) )F )4 )F ) ) )  ,+) t"##J J JF J4 JF J J J  $#J t}) )6 )$ )6 ) ) )  ) t$%%) )6 )$ )6 ) ) )  &%) t&.// &  T    0/" 	$,d.M.QR  .f .6 .f . . .  .d t!)** &     +* )143L3PQRRT8V$$ 	  
  	
 666!"   %$ SR& %-t/D/HIJJ ' ' ''' '
 ' ' ' '  KJ'T &(:;<<S#s/3   f  fff>T8U     =<4 (0$2J2NOPPT8V$$ 	    	
 666!"   %$ QPD $,d.B.FGHH 4 4 444 4
 4 4 4 4 4  IH4n t~S#s 	$ $$$ $ 	$
 666!"$ $ $  $P tTz!2    * &(:;<<S# f C ffn8M     =<4 $,d.B.GHIIV[$11'v '%(F"G ' ' ' 21 JI'$ t'((   	" """ " SM	" " " )("J.
.
. 49d3i . . . .".
.
. 3-. 66>	. . . .(f V      t$%%
 ##!7 7 777 	7
 7 V7 	7 V7 6
7 6666)*7 7 7 &%7t ,4d6R6VWXX     	
   
&	    YX2 t$%%S#4((( +( +(
+(+( +( 	+(
 +( 66>+( +( +( )( &%+(^ t'((
 
 )(
 tz
 W WW	W W 	W
 W W W W  Wt  >#; #; #;L t$%%= =  &%= t%&&> >  '&>( ( (< t-..\S S  /.S t.//\T T  0/T2E 2E 2Ej t$%%= =  &%= t%&&> >  '&> &.&1'/'2	  \& &  &:<G <G <G~ t$%%= =  &%= t%&&> >  '&> &.&1'/'2	  \$( $(  $(N t"##

 

f 

 

v 

 

 

  $#

 t#$$Pv PV P Pf PQW P P P  %$P $dl&6788/0 ' ' ' '  98': &(:;<<&* I I I I  =<I
 t$%%   &% t~  I I I ! I
 $dl&6788" " "  98" t3;<< * * * =<*. tw	 	  	B B B
 
 
 
* 7;j j,jLj $s)S.!j 49c>"	j
 DIsN#j j j U49c>23j j j jZQ Q Q t%-.."$,"$L"$ 5<
 "$ 5<(	"$
 %,'"$ "$ !&"$ "$ "$ "$ /."$J t'((),)L) ,) I	)
 #Y) 3i) ) I) ) ) ) )()X 	8 C@9>9N9N&&: :6 ]59#:BCC  DC, ]59#5=>>S S ?>S
 x 	:?-:O:O66;
 ;
7 
uy}0	1	1	 	 
2	1	
 :?9N9N&&: :6 ]59#5=>>]59#3;<<0 0 =< ?>0d ]59#5<==  >=6 ]59#5=>>]59#5<==  >= ?>8 ]59#5<==]59#5CDD! ! ED >=!F ]59#7?@@]59#<DEE	 	 FE A@	 =BM<Q<QVV= =9 ]59&122 
 
 
 32
8 ]59&?@@@ @ A@@(   t&'' M M M ('Mb(< (< (<X t'/00E E 10EP t UJ UJ UJ   UJp t'((\K( K(  )(K(\ t(011  21" t(011@ @ 21@ t19::F F ;:F, t122\P P  32P
	
6 	
S 	
 	
 	
 	
 t'((UI+ +  )(+\ t011\H H  21H$ t'((UI' '  )('T t011\( (  21(
 t%,--* * * .-* $dl&6788   98 #+T-@-DEFF46 @ @ @c @ @ @  GF@ 	&.	0F0JKLL   ML" 
!4#5#<=>>  ?>D )1233H H 43H: #T[_566./q ' ' ' '  76'0 !()**' ' +*' !)4+=+EFGG  !
 
 
 
 HG
2  ()**  !! ! ! ! +*!H ~B B  B* 0122  32& 89::  ;: ()**@ @ +*@ 0122< < 32< >?@@< < A@<"3 "3 "3 " " " "j j jZ 0122D!)&!1D D D 32D0 +,--; ; .-;  ()**< < +*< t"*++ &  & ,+ &F t#$$G G  %$G* t"*++
 	
`5 `5 `5 ,+`5F t/7885 5 985
 
 
 
 #T[_566=$ = = = =  76= #T^%;<==) ) >=) !			 	 Xy!!   "!	 	 t ())  *) t{"##' ' $#'& tz!""  #" 				! !
 
 
 
 					 * * * *Z      
"DJ$7899  :9
 
 
" &(>?@@  A@ &(>?@@  A@ ty !!& & "!& 
!4:#4566  76 	 $)"2344" " 54" tz!""  #" t~$%% F     &% %&''RV" " "  ('" 0122RV; ; ; 32; &'((KO   )( &(>(FGHH" " " IH" t '((  )(
 t!)**  +* t#$$	 	 %$	 t"##6 6  $#6 t+,,! ! -,! t&''   ('& & & &R tx  5 5 ! 5 tx~J J J    6  ; ; ;h #-Y YYY 	Y 		Y
 	Y 	Y 	Y 	Y 	Y 	Y 	Y Y Y Y Y  !Y" #Y$ %Y& 'Y( )Y* +Y, -Y Y Y Yx;4 ;4 ;4|383838 38 		38
 	38 	38 	38 	38 	38 	38 	38 	38 38 38 38  !38" #38$ %38& '38 38 38 38lI2 I2 I2X t4<==( ( >=(V t+344 # # # 54#L t)122Q Q 32Qh t+,,UI d d d  -,dN t455\b b  65bJ%
V %
6 %
 %
 %
 %
P    t    v 3    $ t,455# # 65#$ t#$$8 8  %$8" t,--\;''! ! (' .-!, 	!"##. . $#. t&'' ) ) ) (')X 	!49=122      32( 
"DJN344      54( t"*++. . ,+. t!)**. . . +*.
	 	 	C 	d 	 	 	 	/ / /
    t{"##' ' ' $#'   6
 
 
 
   
-
 -
 -
 -
b4 4 4 4 t'((& & )(&
 t !!  "!
 !	  & & &  & "	     89::
 #!= ==	= = 	=
 = = E?= = = ;:=@S#X   ( 89:: #!( ((	( ( 	(
 ( ( ( ( E?( ( ( ;:(V EFGG
 #'#!' ''	' ' 	'
 ' ' ' E?' ' ' HG'T 9 ( "" """ 
" 	"
 
" " " " " " " " " " E?" " " 
". 8  "&! 	  	
   E?   
: A  #'!!" !"!"!" 
!" 	!"
 
!" !" !" !" !" E?!" !" !" 
!"H ?@AA
 #'%)!3& 3&3&	3& 3& 	3&
 3& 3& 6"3& E?3& 66>3& 3& 3& BA3&l <=>> !)) ))))	)) )) 	))
 )) )) E?)) )) )) ?>))X = " !4- 4-4-4- 
4- 	4-
 4- 
4- 4- 4- 4- 4- $Z4- 4- E?4- 4- 4- 
4-n 9 * "!" """ 
" 	"
 
" " " " " " " " " " "  E?!" " " 
"0 %   "&*'+"&%)H HH	H H 	H
 H H H H H H E?H smH  }H H 6"H H H 
HV & ( "&*'+#, ,,, 
, 	,
 
, , , , , , , , , , E?,  sm!,"  }#, , , 
,4 )    %!(,!%!%,S ,S,S	,S ,S 6
	,S
 6",S 6",S 3-,S 3-,S ,S ,S ,S E?,S f%,S v,S #,S ,S ,S 
,S^ * * "$("'%47 474747 
47 	47
 6
47 6"47 6"47 ,47 ,47 47 47 47 47 47 47  E?!47" SM#47$  %47 47 47 
47n '()) $(+/'+ _Y _Y
,_Y
,_Y \_Y \	_Y
 5<
 _Y 5<(_Y $_Y _Y _Y _Y *)_YD #')<)DEFF& & &  GF&
 t#'((   )(
  ($*:*>?@@	 	 	 	 	  A@	  , , ,* 	$d&D&LM 
 
 
 
 	$d&D&LM    . (0/7  !% $ %U\ 123 sEL012 uo	
 uo   : 	$d&D&LM 
 
 
 
 					    &$ $ $N t)122
    32 t&''4/ 4/ ('4/n t$,--$% $% .-$%N

 

 


 
 
 #T[%89::4 4 4 ;:4 t!)**   +* ty !!
Q 
Q 
Q "!
Q t,--LP    .-  %t}';<==Xy!!K K K "! >=K  #("9 Q Q Q t7?@@	* 	* A@	* t#+,,	0 	0 -,	0 t!)**  +*> t-566F F 76FD %t~'@ABB27u      CB 
|E E E  E4 	!)4+F+NO     & 5=>??    @?8 t>FGG  HG$ ')<=>>" " "  ?>" uy~())  *)8 uy~ !!6 6 "!6$ t !!
 
	-K -K -K -K  "!-K`   t,--  .- t+,, '
 '
 '
 -,'
T t122    32, t>??    @?. ty8=e 8 8 8 8  8  t~6c 6 6F 6 6 6  6 t'((I& I I I )(I ty"v "& " " "  " t|6& 6V 6 6 6  6' ' '^ "!+/'+ TK TKTKTK el#TK el#	TK
 6
TK 6
TK 5<(TK $TK TK TK TK TKn t   "!'+  6
 6
	
 $     ! & '()) $(#'+/'+  << \ \	
 5<
  5<
  5<( $    *)0 t}	v 	C 	 	 	 	 	  	 t#$$3 3 3  %$3l t~ $8 888 8 	8
 8 8 8 8  8, t3;<<
 	* **&\* c* 	* * * =<*       d2 3 3 3  d4 5 5 5  d4 5 5 5  d= > > >  d= > > >  d= > > >  d= > > >  dD E E E  dD E E E  tB C C C  tB C C C  tB C C C  tB C C C  tJ K K K  tJ K K K  tJ K K K  tJ K K K  t@ A A A  tA B B B  tA B B B  tA B B B   ty   $ t|./       t|./      , 	ty))!!$,//!!$,//
                BN BN BNJ r8   